GAIT TRAINING USING A COMBINATION OF REHABILITATION THERAPIES AND FEEDBACK

A method, apparatus, or system for reducing gait asymmetry is disclosed. The method, apparatus, or system may include measuring one or more gait parameters of an individual; calculating a joint metric based on the measured one or more gait parameters; and signaling an adjustment of an amount of one or more perturbation techniques based on the joint metric for reducing asymmetry in a gait pattern of the individual. The one or more perturbation techniques include asymmetric rhythmic auditory cueing making a first audible cue with a first cue duration for one leg of the individual and a second audible cue with a second cue duration for another leg of the individual. Here, the first cue duration is different from the second cue duration. Other aspects, embodiments, and features are also claimed and described.

Skip to: Description  ·  Claims  · Patent History  ·  Patent History
Description
CROSS REFERENCE TO RELATED APPLICATION

This application claims the benefit of U.S. Provisional Patent Application Ser. No. 63/105,761, filed Oct. 26, 2020, and U.S. Provisional Patent Application Ser. No. 63/105,759 filed Oct. 26, 2020, the disclosures of each of which are hereby incorporated by reference in their entirety, including any figures, tables, or drawings.

STATEMENT OF GOVERNMENT SUPPORT

This invention was made with government support under 1910434 awarded by the National Science Foundation. The Government has certain rights in the invention.

BACKGROUND

Walking is fundamental to independence, and improving walking function is the goal most frequently voiced by stroke survivors. Despite this, only 7-22% of people with a stroke are able to regain sufficient function to be considered independent community ambulators. Also, people suffering from conditions affecting lower limbs experience diminished abilities to walk or to maintain a consistent gait. For example, people living with a partial or complete loss of a lower limb experience limited abilities to walk and run. There is, therefore, a significant need to improve systems and methods of improving the gaits of individuals with diminished abilities to walk or to maintain a consistent gait.

SUMMARY

The following presents a simplified summary of one or more aspects of the present disclosure, in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated features of the disclosure, and is intended neither to identify key or critical elements of all aspects of the disclosure nor to delineate the scope of any or all aspects of the disclosure. Its sole purpose is to present some concepts of one or more aspects of the disclosure in a simplified form as a prelude to the more detailed description that is presented later.

In one example, a method, apparatus, or system for reducing gait asymmetry is disclosed. The method, apparatus, or system may include measuring one or more gait parameters of an individual; calculating a joint metric based on the measured one or more gait parameters; and signaling an adjustment of an amount of one or more perturbation techniques based on the joint metric for reducing asymmetry in a gait pattern of the individual. The one or more perturbation techniques include asymmetric rhythmic auditory cueing making a first audible cue with a first cue duration for one leg of the individual and a second audible cue with a second cue duration for another leg of the individual. Here, the first cue duration is different from the second cue duration.

In another example, a method, apparatus, or system for reducing gait asymmetry is disclosed. The method, apparatus, or system may include measuring a left step length and a right step length; measuring an average step length for a predetermined period of time; calculating a left leg long percentage based on the left step length, the right step length, and the average step length; comparing the left leg long percentage with a predetermined left leg long percentage; and signaling an adjustment of amounts of speeds based on the compared left leg long percentage.

These and other aspects of the invention will become more fully understood upon a review of the detailed description, which follows. Other aspects, features, and embodiments of the present invention will become apparent to those of ordinary skill in the art, upon reviewing the following description of specific, exemplary embodiments of the present invention in conjunction with the accompanying figures. While features of the present invention may be discussed relative to certain embodiments and figures below, all embodiments of the present invention can include one or more of the advantageous features discussed herein. In other words, while one or more embodiments may be discussed as having certain advantageous features, one or more of such features may also be used in accordance with the various embodiments of the invention discussed herein. In similar fashion, while exemplary embodiments may be discussed below as device, system, or method embodiments it should be understood that such exemplary embodiments can be implemented in various devices, systems, and methods.

BRIEF DESCRIPTION OF THE DRAWINGS

Many aspects of the present disclosure can be better understood with reference to the following drawings. The components in the drawings are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles of the present disclosure. Moreover, in the drawings, like reference numerals designate corresponding parts throughout the several views.

FIG. 1 depicts components and objectives for an exemplary method of controlling gait rehabilitation therapies in accordance with various embodiments of the present disclosure.

FIG. 2 represents a step length as corresponding to the distance between feet at heel strike.

FIGS. 3A-3B are charts illustrating that two physically different systems (FIG. 3A) can generate the same motion, but will have different forces; likewise, an applied force (FIG. 3B) will lead to different motions.

FIGS. 4A-4B are graphs showing how height and weight combine to impact step time and step length.

FIG. 5 is a chart showing how gait parameters of a model with asymmetric simulated spasticity change as a function of asymmetric roll-over shapes.

FIG. 6 is a graph of preliminary data showing the split-belt treadmill (SBT) congruently combined with asymmetric rhythmic auditory cueing (ARAC), wherein each parameter in ARAC & SBT is approximately the sum of each therapy applied individually.

FIG. 7 is a representation of a multidimensional gait parameter space in accordance with various embodiments of the present disclosure.

FIG. 8 is a block diagram of an exemplary feedback control system used to optimize a gait pattern based on a joint metric of many gait parameters in accordance with various embodiments of the present disclosure.

FIG. 9 shows a conventional technique for collecting gait perception ratings from a gait simulation.

FIG. 10 shows box and whiskers notch plots of participants' responses to gait videos (showing a combination of asymmetric and symmetric Step Length and Swing Time features) in accordance with various embodiments of the present disclosure.

FIG. 11 shows the predicted behavior of the neuromusculoskeletal system by summing gait responses of two or more stimuli.

FIG. 12A shows examples of types of therapeutic interventions with symmetric and asymmetric proportions to the left and right legs. FIG. 12B shows an example of an experimental design of four trials in an experiment. FIG. 12C shows an example of a timeline of a duration of each section in the experiment.

FIG. 13A shows percent asymmetry averages of step length, step time, and peak vertical force during five time-windows for trial 1. FIG. 13B shows percent asymmetry averages of step length, step time, and peak vertical force during five time-windows for trial 2. FIG. 13C shows percent asymmetry averages of step length, step time, and peak vertical force during five time-windows for trial 3. FIG. 13D shows percent asymmetry averages of step length, step time, and peak vertical force during five time-windows for trial 4.

FIG. 14A shows an example of a linear model for a group having a coefficient value predicting the asymmetric performance of step length, step time, and peak vertical force compared to real data during trial 3 and 4. FIG. 14B shows an example of a linear model for a group having a different coefficient value predicting the asymmetric performance of step length, step time, and peak vertical force compared to real data during trial 3 and 4.

FIG. 15 shows an example of a whisker plot of coefficients for individual lineal models.

FIG. 16 shows an example of a block diagram of real-time feedback algorithm by a real-time feedback controller.

FIG. 17 shows an example of an experimental protocol in table format.

FIG. 18 shows exemplary markers placed on a participant.

FIG. 19 illustrates an example of a linear model used for second-session target asymmetry prediction.

FIG. 20 shows an outcome tree with categorized participant results.

FIG. 21 shows Trial 2 controlled intervention examples from the most responsive participants.

FIG. 22 shows Trial 2 maximum intervention examples featuring the standard insufficient symmetry achievement and the sufficient symmetry outlier.

FIG. 23 shows Trial 1 belt speeds for participants in unsuccessful correlation, successful correlation—maximum intervention, and success correlation—controlled intervention categories.

FIG. 24 depicts a schematic block diagram of a computing device that can be used to implement various embodiments of the present disclosure.

FIG. 25 is a flow chart illustrating an exemplary process for reducing gait asymmetry according to some embodiments.

FIG. 26 is a flow chart illustrating an exemplary process for reducing gait asymmetry according to some embodiments.

 DETAILED DESCRIPTION

In accordance with the present disclosure, embodiments of systems and methods for controlling gait rehabilitation therapies are presented. The present disclosure posits that a combination of therapies can generate balanced gait patterns such that the overall symmetry of gait parameters improves. The present disclosure tests the assumption that a completely symmetric gait is ideal (or even possible) for a person that inherently has an asymmetric impairment. Thus, allowing small asymmetries in some parameters can reduce the overall asymmetry.

A diffuse neurological injury, like stroke, often affects many different aspects of motor control and each affected individual presents different symptoms. Even among people who have hemiparesis (i.e., unilateral weakness) affecting the lower limbs, there is wide variability in the type and magnitude of resultant gait impairments. Clinical physical therapists are faced with the challenge of how to tackle these impairments: should effective gait rehabilitation focus on whole-body movements (i.e., walking) without precise corrections of person-specific deficits, or should approaches aim to correct specific gait coordination deficits? Current clinical practice and many scientific studies of gait rehabilitation take a middle-ground approach, which involves whole-body walking practice in conjunction with one or more impairment-specific interventions. While some of these interventions result in positive outcomes, the critical ingredients of a successful training intervention remain unknown. Moreover, although multiple gait impairments are often targeted in different ways over the course of a walking training intervention, the interactions between different treatment modalities are under-acknowledged and understudied.

1. Examples of Feedback-Driven and Customized Gait Rehabilitation Systems and Methods 1.1. General Concepts

FIG. 1 shows an example of a combined-therapy concept, in accordance with various embodiments of the present disclosure, that focuses on systems and methods that can select from, and employ one or multiple different interventions to correct gait asymmetry. In particular, the figure illustrates why multiple therapies used together are needed to generate a customized gait pattern. In some instances, using one type of interventional therapy will suffice. In other instances, however, one therapy alone will reduce the asymmetry, but may not have the desired effect on all parameters. In accordance with the present disclosure, additional therapies can, if chosen correctly, affect the parameters that the first therapy did not affect in the desired way. That is, multiple therapies apply simultaneous stimuli to affect a wider range of gait parameters and create flexible training regiments. If multiple therapies span the gait parameter space in terms of their effect on the parameters, feedback can be applied to enable a more optimized and individualized gait pattern. Previous methods for providing therapy were either entirely manual (relying on the instincts and observations of a therapist), based on trial and error (e.g., application of one therapy after another, until long term success is achieved), or both. Now, the present disclosure provides systems and methods for combining one or more therapies with real-time feedback to reduce gait asymmetry.

The four first (“b”) bars on the left represent individual gait parameter asymmetries. The second (“o”) bar on the right represents a combined metric of their asymmetry. Therapy 1 generates one gait pattern while Therapy 2 generates a different gait pattern; both therapies improve the overall symmetry, but in linearly independent ways. Utilizing the systems and methods herein, various embodiments can select an appropriate combination of the right therapies, such as applying a smaller dose of Therapies 1 and 2 together. By doing so, the gait pattern can be corrected more so than either does independently, though this may not be readily apparent to a human therapist who is performing therapeutic intervention manually. Adding additional therapies allows further control over each individual gait parameter in a measurable way that can be tracked and updated in real time using data not necessarily readily observable by human therapists. To achieve the desired gait parameters, a real-time gait assessment and controller (“feedback controller”) can adjust the amount of each therapy applied.

Additionally, the present disclosure explores identifying how real-time feedback methods can be used to restore balanced gait patterns in people with asymmetric walking patterns and how a person can best interact with multiple therapies applied at the same time. Although the present disclosure focuses on interlimb gait parameters (i.e., symmetry), such methods can work with any gait parameter that can be measured and affected by a rehabilitation therapy. For example, problems with coordination across joints within a limb (intralimb coordination) that contribute to foot drop during swing or weak push-off in late stance can be targeted with a similar approach.

1.1.1 Definitions

The following are definitions of phrases that are provided for clarity and convenience: Gait parameters—Quantifiable measures of the temporal (e.g., stance/swing time) and spatial (e.g., step length) aspects of a gait, the kinematic joint angles, the resulting ground reaction forces, and the forces/torques on a person's joints; Physical properties—Any bodily characteristic affecting gait parameters (e.g., mass, neural control); Physical perturbations/therapies—Changes to physical properties or external manipulations that affect gait parameters (e.g., adding weight to the leg, pulling on the hip (e.g., via an elastic band, belt, cable, etc.), or sounds indicating when to step); Complete symmetry—Gait that is symmetric between both legs in all gait parameters; Balanced gait patterns—An unimpaired gait that is slightly asymmetric in some gait parameters to balance the asymmetries that will develop in other gait parameters.

1.1.2. Scientific Challenges

One of the scientific challenges associated with the present disclosure is determining what the ideal gait pattern is for a specific individual with an asymmetric impairment. The present disclosure posits that symmetry-based rehabilitation methods post-stroke are only partially effective because an asymmetric impairment prevents a person from returning to a completely symmetric gait. In contrast to many conventional methods, an exemplary technique, in accordance with the present disclosure, embraces the inherent asymmetry in the person.

As an example of some types of asymmetric gaits, swearing a thick-soled or heavy shoe on one foot while barefoot on the other; while trying to walk with symmetric step timings, step lengths, and ground reaction forces. This simple physical perturbation will asymmetrically modify one's gait in at least one of the three measures (of step timings, step lengths, and ground reaction forces). Walking with such a physical perturbation will also change one's gait pattern to such an extent that other people are likely to perceive a difference. In contrast to the one physical perturbation (e.g., heavy shoe), in this example, the millions of individuals with a stroke each have multiple asymmetric changes, thus compounding the effect, and many of them would gain additional mobility, confidence, and independence from rehabilitation methods that focus on balancing all aspects of a gait pattern.

Another scientific challenge is to understand how feedback can be used to control multiple asymmetric physical changes to beneficially affect the underlying gait. Unlike existing methods, an exemplary technique, in accordance with the present disclosure, incorporates the existing gait asymmetry and aims to find a functional balance between the perception of gait and the kinematic and kinetic measures. Although the resulting gait will have some degree of asymmetry in all measures (i.e., balanced gait patterns), it will more likely meet the functional walking goals of individuals with asymmetric impairments. These balanced gait patterns will include small deviations from complete symmetry to globally optimize and improve the overall dynamics, kinematics, kinetics, and perception of gait.

1.1.3. Control of Multiple Therapies

The present disclosure further posits that controlling multiple therapies simultaneously will allow individual gait parameters to be customized in a beneficial way that is not possible with a single therapy. The systematic approach to finding globally-balanced gait patterns proposed here incorporates a model of how gait therapies combine, psychological gait-perception experiments, and human gait studies using feedback. The results will provide fundamental insights into gait and clinical rehabilitation based on multiple feedback methods by jointly analyzing how asymmetric impairments affect gait dynamics.

Since many physical therapy clinics neither have the time nor funds to analyze a patient's gait pattern using motion capture, the physical therapists rely on their perception of the patient's gait for assessment. However, it is not clear how well gait observations relate to gait function, much like how the measured gait parameters in the literature do not necessarily correlate to gait function. The present disclosure posits that trained physical therapists will detect subtler deviations from symmetry than untrained people and that both groups will perceive measures of asymmetry differently, even if the parameters have the same magnitude of change. To test these hypotheses, physical therapists as well as untrained healthy and impaired individuals will rate how various gait patterns appear. The results will help us understand what clinical physical therapists perceive about gait and help direct their attention to important parameters, particularly those that both have a large impact on gait function and are not easily noticed.

The present disclosure models how rehabilitation methods combine and interact when simultaneously applied to a person. In one non-limiting example, the present disclosure examines how the following external perturbations jointly affect gait: adding both (1) weight and (2) height to the foot, (3) lateral pulling at the hip combined with (4) anterior pulling on a foot, and a (5) split-belt treadmill combined with (6) asymmetric rhythmic auditory cueing. Using real-time feedback (based on multiple gait parameters) to control multiple therapies, the six external perturbations (noted above) can be combined into one system that uses feedback to alter the magnitude of each one so as to minimize the overall asymmetry of the gait pattern, in accordance with various embodiments of the present disclosure. To help understand how people perceive gait patterns, a subset of the collected gait patterns can be added to an online repository where people can rate the gait patterns. This will allow for the perception of gait to be correlated to the underlying gait parameters. The perception could then be used as an additional parameter to evaluate what level of asymmetry is acceptable for an individual. It is noted that techniques of the present disclosure are not limited to the illustrative examples disclosed herein and may be applied to a variety of therapies and a variety of combination of therapies.

1.2. General Concepts

An example of a common gait impairment following stroke is hemiparetic (or asymmetric) gait, resembling a limp. Hemiparetic gait results from weakness and poor range of motion in the paretic (more affected) leg, limiting leg extension at the end of stance phase and is correlated with decreased walking speed, reduced walking efficiency, increased joint and bodily degradation, and increased susceptibility to injuries and falls. Since the paretic leg does not extend backward, the person can only take a small step forward with the non-paretic leg. People with stroke may compensate with larger paretic-side steps, leading to step length asymmetry (as illustrated in FIG. 2, in which step length corresponds to the distance between feet at heel strike, such that poor propulsive force generation by the paretic leg (black) leads to a small nonparetic (grey) step), a key feature of hemiparetic gait. Many recent studies have investigated rehabilitation approaches that fix this “problem” of gait asymmetry poststroke, yet it is not clear if gait symmetry is ideal or even achievable for a person who has two limbs with fundamentally asymmetric properties. One objective of the present disclosure is to minimize the overall asymmetry by taking into account multiple gait parameters including how the gait pattern appears.

Outcomes of gait rehabilitation can be quantified as changes in walking function (e.g., how well can a person walk) or walking quality (e.g., how well-coordinated is the gait pattern). A common measure of walking function is walking velocity. Walking velocity is indicative of overall gait performance and can differentiate levels of disability; a velocity of 0.8 m/s is often considered the minimum threshold for community ambulation, and people who are able to ambulate in a shopping center have a mean gait speed of 1.14 m/s. A common measure of walking quality is gait symmetry. Normal gait tends to be symmetric in kinematics and dynamics and has a 4-6% difference in vertical force and temporal parameters between the two legs. Gait often becomes hemiparetic as a consequence of unilateral stroke affecting leg motor areas. Hemiparetic gait is characterized by significant asymmetry in temporal (e.g., double-limb support time) and spatial (e.g., step length) measures of interlimb coordination. Propulsive force of the paretic limb is reduced compared to the nonparetic limb, as are work and power of the paretic plantar flexors. Vertical ground reaction forces are decreased on the paretic limb relative to the nonparetic limb, reflecting diminished paretic limb weight bearing.

Although both function (e.g., velocity) and quality (e.g., symmetry) are frequently used to evaluate the effectiveness of gait training interventions, only a few studies have specifically examined the relationship between the two, and these studies often report conflicting findings. While one study found no relationship between temporal symmetry and gait velocity in people with hemiparetic gait, another study found that preferred velocity was negatively associated with temporal asymmetry. The study also reported that spatial asymmetry was not related to gait velocity, but a different study noted a negative correlation between the two (albeit a weak one: r=−0.35). The studies have argued that asymmetries in propulsive forces generated by each leg are correlated with hemiparetic severity and step length asymmetry, but there was little support for a direct relationship between asymmetric propulsive forces and gait speed. Interestingly, a study found that gait velocity continues to improve in the months and years following a stroke, but gait symmetry (temporal and spatial) worsens over this same timeline. A different study also reported the surprising finding that people with stroke are kinematically more symmetric while walking on a treadmill compared to over the ground, but the metabolic cost was also increased during the (more symmetric) treadmill walking. Overall, these studies indicate that the relationship between gait quality and gait function remains unclear. A more comprehensive approach taking into account the interactions between kinematics, dynamics, perception, and gait function is necessary.

1.3. Preliminary Experiments

1.3.1 Assumption that Symmetry is Ideal (Computational Perspective)

People with hemiparesis due to stroke have different force and motion capabilities on each leg. The paretic leg is weaker and has a more limited range of motion than the non-paretic leg. Rehabilitation science has not advanced to the point where these problems can be fully corrected. Therefore, when we are retraining walking post-stroke, we are working with an inherently asymmetric system. From a biomechanical view, two physically different systems (e.g., legs) can only have the same motion if the forces controlling them or the forces resulting from the movement are different. When an individual with an asymmetric impairment walks with symmetric step lengths, other aspects of gait become asymmetric, such as the forces in the joints, the amount of time standing on each leg, and other temporal variables, all of which can be detrimental to efficiency and long-term viability. Understanding how symmetry affects function could change the fundamental nature of clinical gait rehabilitation. To demonstrate the tradeoffs between different asymmetries, a computational model of walking-like dynamics is used in various embodiments.

The physical properties of legs during walking are often modeled as inverted pendulums. Here, a double-link pendulum is used to demonstrate symmetric motion in an asymmetric system. Each pendulum can be thought of as a leg in an asymmetrically impaired individual. In various embodiments, two double-link pendulums are created with different masses and mass distributions, such that the elements of the matrices that define the dynamics are equal. This leads to identical free-swinging motions. By modeling two masses on each link, two physically dissimilar pendulums can be made to have the same motion. Typically, one mass per link is used, but the extra mass makes the dynamics equations underconstrained, so there are an infinite number of four-mass double-link pendulums with the same free-swinging motions. This shows that kinematic symmetry alone is possible, but that is only one part of the human gait.

To understand the influence of forces and heel strike, the pendulum dynamics were simulated with an inelastic collision applied to the end of link one at 0.56 s. FIGS. 3A-3B show the link angles and forces applied at the pivot point. The free motion is the same before collision (i.e., symmetric motion), but the forces are different. After the collision, which simulates walking impacts (e.g., heel touch), the motions no longer match. The different masses and mass distributions lead to different kinetic energies, impact forces, and, thus, different resulting motions after impact. Thus, restoring kinematic symmetry in humans will likely be at the expense of kinetic symmetry, or vice versa. This suggests that an asymmetric person cannot walk completely symmetrically. Much of the stroke rehabilitation literature focuses on trying to obtain symmetry in one or a few gait variables. In contrast, the present disclosure examines how symmetric an asymmetric person can become.

1.3.2. Combined Height and Weight Effects

Preliminary experiments have been performed to examine how two external physical perturbations (added height and mass) applied on a person's leg jointly affect gait. Past studies have shown how height independently affect gait patterns.

Experimental procedure: Baseline (no height or weight) was tested first and then a random order of all other combinations (height: none, small, & big crossed with weight: none, small, & big; includes height and weight on the same and opposite feet). The treadmill speed was set to the subject's comfortable walking speed calculated using the average of three 10 meter walk tests. The subjects walked for two minutes in each height/weight configuration and the last minute was used for gait parameter measurement. In particular, twenty subjects walked with all 15 combinations of added masses (4.6 kg, 2.3 kg, or 0.0 kg) and heights (5.2 cm, 2.6 cm, or 0.0 cm) including added on the same and different sides. An additional baseline trial was performed last; no statistically significant differences were found between the first and last baseline trial. Twenty subjects were tested on the CAREN (Computer Assisted Rehabilitation Environment) system using a data analysis procedure for data collected from force plates and motion capture.

Results: A two-way ANOVA (Analysis of Variance) with interaction effects using weight and height as independent measures was performed for each of the gait parameters, as shown in Table 1 (below). Height and weight were statistically significantly different for all cases except leg length's effect on push-off force. In all cases, there were no statistically significant interaction effects between mass and leg length. The lack of interaction suggests that each variable acts independently at the magnitudes tested. This independence could make it easier for the feedback controller (using feedback of gait parameters for control of multiple therapies) to minimize the asymmetry of multiple gait parameters.

TABLE 1 Measure Mass Effect Height Effect Interaction Effect Step Time p < .0001 p < .0001 p = .08 F(4, 165) = 87.5 F(2, 165) = 56.0 F(8, 165) = 1.8 Step Length p < .05 p < .0001 p = .97 F(4, 165) = 2.9 F(2, 165) = 15.3 F(8, 165) = 0.29 Peak Vertical Force p < .0001 p < .0001 p = .84 F(4, 165) = 9.6 F(2, 165) = 22.6 F(8, 165) = 0.53 Peak Pushoff Force p <.0001 p = .26 p = .18 F(4, 165) = 137.0 F(2, 165) = 1.35 F(8, 165) = 1.44 Peak Braking Force p < .0001 p < .05 p = .73 F(4, 165) = 137.0 F(2, 165) = 4.1 F(8, 165) = 0.66

FIGS. 4A-4B show the changes to gait parameters from the 15 combinations of height/weight and the resulting statistics. The figures show how height and weight combine to impact step time and step length. Error bars show standard error. The ★ shows a configuration where asymmetric physical alterations lead to a symmetric gait parameter, but the others are still asymmetric (e.g., at 4.6 kg and 2.6 cm, step length is symmetric, but step time is asymmetric).

There is a configuration where step length is symmetric even though the added height and weight are both generating a physical perturbation (shown as ★ in FIGS. 4A-4B). The two external physical perturbations (e.g., added height and added weight) are essentially canceling each other out. However, none of the other parameters are the same at that point since the effect of height and weight is unique on different measures of gait.

1.3.3. Therapy Incorporating Height, Weight, and Motion (Gait Enhancing Mobile Shoe)

An example of a stroke rehabilitation therapy that incorporates a change in height, added mass, and a motion applied to the foot is the Gait Enhancing Mobile Shoe (GEMS). It functions by moving the healthy foot backward while the individual walks over ground. The backward motion is generated passively by redirecting the downward force when the user steps onto it, which requires a slight change in height. The shoe also has mass, so the height and weight effects are combined along with the foot moving backward during stance. Results from the registered clinical trial with six subjects each using the GEMS for twelve 30-minute sessions demonstrated significant improvements between pre- and post-training in step length symmetry, double limb support symmetry, gait velocity, and the Timed Up and Go test.

After-effect mechanism: The GEMS helps with gait relearning by exaggerating the gait asymmetry such that the resulting gait pattern is more symmetric once the shoe is taken off. This after-effect occurs in the opposite direction and corrects the user's gait once the shoe is removed. Note that the shoe could also be used to compensate for the asymmetry (by placing it on the paretic foot) such that the user walks symmetrically while wearing the device, but this provides no benefit once the shoe is removed.

1.3.4. Role of Perception in Gait

Walking asymmetry is detrimental to the body, but also affects how the person appears. FIG. 5 shows how ground reaction forces, step time, and step lengths change as a function of an asymmetric physical perturbation (i.e., roll-over shape in this case) as well as how the perception of that gait pattern changes (right axis). For the preliminary experiment, a passive dynamic walker (PDW) model, which is a completely passive system that can walk using only gravitational energy and has been used to understand the passive mechanics of walking separate from a control system, was used. The PDW model used in this example has simulated spasticity at one of the knee joints and different roll-over shapes on the feet. The roll-over shape asymmetry is the percentage difference between the radius of one foot bottom shape compared to the other foot bottom shape ranging from 0% (same radius) to 120% (more than two times the radius). The spasticity causes an asymmetry even when the roll-over shapes are the same (i.e., 0%), and changing one of the rollover shapes creates an asymmetry that further affects the gait pattern.

No two gait parameters are simultaneously symmetric given these changes in roll-over shape. However, allowing all of them to be slightly asymmetric, such as near 80% in FIG. 5, minimizes a combination of all the asymmetries and the gait pattern is still perceived as mostly normal. It is generally hard to agree on an ideal gait pattern on this graph. However, many studies focus on step length and would attempt to achieve the gait pattern near 100% even though step time is larger, and appearance begins to be noticeable.

1.4. Examples

In the present disclosure, multiple different therapies (e.g., 6 therapies initially) can be used to change a user's gait. The choice of methods or therapies is largely based on how they physically interact and the different mechanisms by which they interact. The six physical perturbations that are examined in the present proof-of-concept study are: 1) Split-Belt Treadmill (SBT); 2) Asymmetric Rhythmic Auditory Cueing (ARAC); 3) Lateral Pulling at the hip; 4) Anterior Pulling on a foot; 5) Adding Weight to a foot; and 6) Adding Height to a foot. These methods do not physically interfere with each other, and they all have distinctly different effects on the gait. The first four can be modified in real-time based on feedback measured from the gait parameters. The last two cannot be modified in real-time. There are many other methods that could also be used, but these six should allow us to understand how this concept of combined therapies can be scaled to include other methods of changing a gait pattern and to include other counts of therapy methods.

First, split-belt treadmill (SBT) is a treadmill with is a treadmill with two treads that can move at different speeds. It has been used for stroke rehabilitation and predominantly changes step length asymmetry. The SBT uses the after-effect mechanism similar to that generated by the GEMS.

Asymmetric Rhythmic Auditory Cueing (ARAC) involves applying beeps with different time gaps before each step indicating to the subject when each foot should be placed. Symmetric Rhythmic Auditory Cueing has been utilized for stroke rehabilitation and its effectiveness on gait symmetry has been shown in the literature. The asymmetric version presented here has not been demonstrated in the literature, but preliminary results show that it can be adjusted to generate gait changes.

Lateral pulling at the hip (e.g., via an elastic band, belt, cable, etc.) causes the individual to have to overcome a sideways force, which tends to make them lean away from the force and take a shorter step on the foot on the side being pulled.

Anterior pulling on a foot (e.g., via an elastic band, belt, cable, etc.) assists a leg to appropriately swing during walking. Pulling on the paretic foot during treadmill walking significantly improved gait speed and step length.

Adding weight to a foot has been previously discussed herein. Weight cannot be changed in real-time, so the treadmill can be stopped while this change occurs.

Adding height to a foot has been previously discussed herein. Height cannot be changed in real-time, so the treadmill can be stopped while change occurs.

1.4.1. Model of Interactions Between Multiple Therapies

In order to model some of the interactions between the six methods of gait change and given that examining all 30 interactions is labor intensive, the present disclosure focuses on a subset of the interactions that will give us a baseline for how each of the six methods work individually and in combination with one other. Based on this study, the statistics and results shown in Table 1 (for height/weight study) can be generated for the following three combinations: 1) Height and Weight; 2) Lateral Pulling at the hip and Anterior Pulling on a foot; and 3) Split-belt Treadmill and Asymmetric Rhythmic Auditory Cueing. The above experiments will use similar methods, numbers of subjects, and statistical analyses as the preliminary experiments (previously described herein).

For Height and Weight (as described with respect to the preliminary experiments), the results showed that there was no interaction effect given the height and weight tests. It is not expected that the height or weight used in the preliminary experiments will be exceeded when three or more methods are used together.

For Lateral Pulling at the hip and Anterior Pulling on a foot, these two methods both involve pulling. Thus, we expect the combination of these two to have the highest chance for an interaction effect. If an interaction effect is found, a linear model may not suffice and we can use a nonlinear model to examine the interactions. This may complicate the feedback control design slightly, but would not pose any significant delay.

For Split-belt Treadmill and Asymmetric Rhythmic Auditory Cueing, FIG. 6 shows a preliminary experiment with one subject walking in three conditions: (i) ARAC alone, (ii) SBT alone, and (iii) ARAC & SBT together. The effects of ARAC alone added to the effects of SBT alone are approximately equal to the effects of ARAC & SBT simultaneously applied, which demonstrates that these two methods can be used in conjunction with each other to jointly affect gait patterns. A particularly interesting aspect of this combination is that SBT can have an opposite direction after-effect while ARAC can train the user directly in the desired timing; thus, these two therapies can be applied congruently (faster tread associated with longer step time; as tested here) or incongruently (faster tread associated with shorter step time).

1.4.2. Feedback of Gait Parameters for Control of Multiple Therapies

In accordance with the present disclosure, controlling multiple therapies simultaneously can allow independent control of individual gait parameters. The goal of controlling each parameter is to bring the entire gait closer to symmetry. Since each gait parameter shows a different aspect of a gait pattern, a joint metric or measure of symmetry can be calculated that incorporates all measures of the gait pattern.

Metric of overall symmetry: In one embodiment, the Combined Gait Analysis Metric (CGAM) is used to select a joint metric/measure of symmetry. CGAM is a normalized version of the Mahalanobis distance that is a measure of the distance between a point (perfect symmetry in our case) and a distribution (the measured gait pattern). The Mahalanobis distance is a multidimensional generalization that measures how many standard deviations the point is away from the mean of the distribution, and it can be used to quantify gait patterns for recognizing a person by gait. The Mahalanobis distance can incorporate any measure of gait including a measure of perception, in accordance with various embodiments of the present disclosure.

The example in FIG. 1 shows CGAM as the combined metric on the right of each graph. Perfect symmetry is represented by a magnitude of 0, which is not expected, even in healthy individuals. It is shown a different way in FIG. 7 where CGAM is a single representation of the measured gait parameters that generally scales with the global deviation from symmetry and lets you determine how far away a gait is from an ideal. The deviation of each measure is scaled based on the variance within that measure, so measures that generally have larger magnitudes of asymmetry (e.g., forces) will be scaled so that each gait parameter has a similar influence on the overall metric. These weightings are a starting point and the weight assigned to each metric can be evaluated and can be adjusted based on the effect of each parameter on appearance and based on the biomechanics related to which gait parameters are most important. In the figure, the gray (“o”) lines represent the distance each gait is from a symmetric gait. In one embodiment, the feedback can be based on 11 gait dimensions (i.e., parameters), but that is hard to visualize. Accordingly, FIG. 7 shows 3 gait dimensions.

CGAM was evaluated on data from a clinical trial using the GEMS. CGAM scores calculated from the spatiotemporal measures were compared to functional measures and moderate correlations to the six-minute walk test (R2=0.41), timed up and go (R2=0.22), and gait velocity (R2=0.51) were found. This correlation to functional outcomes indicates that the CGAM is a good metric, among others, to use for the feedback controller to minimize. Other methods of combining multiple gait parameters may also be deployed, such as, but not limited to, Principle Components Analysis, Independent Component Analysis, and machine learning.

Rationale: Instead of making one gait parameter symmetric, in various embodiments, an exemplary method for controlling gait rehabilitation therapies focuses on minimizing a joint metric, such as CGAM, that allows some asymmetry in the gait parameters so that all of them will be relatively small and none will be excessively large. In various embodiments, a feedback controller can measure the gait parameters in real time and then calculate what changes in each of the external perturbations will likely lead to a reduction in the joint metric of asymmetry and a balanced gait pattern. The iterative perturbation changes will be based on the data collected during simultaneous or concurrent therapy techniques being collectively applied, such that the collected data relates to the individual therapy methods and their interactions. The goal is not to identify a new rehabilitation method; rather, the goal is to understand how multiple therapies can be used to optimize a gait pattern. By controlling multiple therapies simultaneously, individual gait parameters are allowed to be customized or adjusted.

Controller design: To allow for minimization of the gait parameter joint metric, multiple controlled inputs are utilized. Although a perturbation input can only change the gait parameters in a certain ratio, each perturbation input affects different parameters uniquely. For example, adding height to one foot increases step length and has a small effect on braking force, but adding weight has a small effect on step length and a large effect on braking force. These two perturbations have different effects that span the gait space differently. Thus, controlling multiple perturbations increases the reachable gait-parameter space and increases the chance of finding a balanced gait pattern for each individual.

In the present non-limiting study, six external perturbations are chosen because they can be independently controlled, have been shown to asymmetrically change gait patterns, and their effect on gait patterns are unique. The six therapies can be thought of as six linearly independent vectors of size 11 that represent the change in each of 11 gait parameters. This leads to an under-actuated optimization problem with uncertainty. To help solve this problem, constraints can be applied to limit each external perturbation.

The 11 gait parameters to be measured include the amount of asymmetry of: spatiotemporal measures (<1> step length, <2> step time, <3> double limb support time, <422 stance time, and <5> swing time); kinematic measures (<6> hip angle, <7> knee angle, and <822 ankle angle); and force measures (<9> vertical force, <10> braking force, and <11> push-off force).

In accordance with various embodiments, FIG. 8 shows a feedback loop (for the feedback controller) that will use the 11 gait parameters measured in real-time to change each of the six therapies to minimize the joint metric of gait parameter asymmetries. In various embodiment, the controller can be designed based on SIMO/MIMO control, adaptive control, machine learning approaches, and/or linear/nonlinear optimization methods to find solutions to the under-actuated problem.

The following is an exemplary, non-limiting, procedure for controlling multiple gait parameters:

1) Measure gait velocity over ground using the 10 m walk test.

2) Have participants walk for two minutes at comfortable speed on a split-belt treadmill with tied belts.

3) Measure gait parameters using force plates in the treadmill and motion capture.

4) The controller will calculate the split-belt ratio, lateral pelvic force, auditory cueing ratio, and anterior pulling force on a foot every minute based on minimizing the overall asymmetry of all gait parameters.

Force/tread/sound ratio will change slowly over ten seconds. The belt speeds will not change by more than 0.05 m/s at each increment. The belt and auditory ratios will not exceed 3:1. The lateral pelvic force will not change by more than 10N at each increment with a maximum force of 100N. The anterior foot force will not change by more than 3N at each increment with a maximum force of 30N.

5) Gait parameters will be continuously measured for one minute, but the controller will use an average of the last twenty seconds to calculate the changes for the next iteration. The experiment operator will be shown what change will be applied with a ten second window to cancel that change for safety.

6) Ask participants if they need a break every three minutes. When restarting after a break, keep the same tread speeds and lateral pelvic force, but start at step 3.

7) Repeat steps 4 through 6 until 15-20 minutes has been reached or a balanced gait pattern has been achieved.

Statistical analysis: During an exemplary gait analysis procedures, changes in the 11 gait parameters at each stage of the iterative process can be analyzed as the controller adapts the gait. A multiple regression analysis can also be performed to determine the effect of each of the six methods of changing gait. The perception of each of these gait patterns will also be measured and compared to the individual gait measures once the perception study is complete.

Patient variability: Stroke can affect each person differently which results in a unique gait pattern. Responses to therapies are generally in similar directions across subjects, but there are cases where the patients respond differently or not at all. Although the controller may be based on generalized responses to each therapy, the feedback will customize the amount of each therapy added to the individual. In the specific cases that the patient is very different than the general trend, the controller may be configured to operate in a static mode such that the feedback is not used to customize or change the amount of therapy that is applied to the individual.

1.4.3. Perception of Gait

In various embodiments, a gait perception metric can be used as feedback to customize a user's therapy exercises. To explain, the “Uncanny Valley” shows how motions that are familiar and close to typical human motions, yet slightly different, can be very noticeable. When a person sprains his/her ankle, the change in gait pattern is quickly apparent to other people. However, the slight asymmetries in healthy individuals are not typically noticeable nor detrimental. Many studies have tried to make simulated motions and robots appear humanlike, but the present disclosure considers where the transition occurs between perceiving motion to be from an able-bodied person or from someone with a gait impairment. A goal is to find two perceptual limits: one specifying the perception of a normal gait pattern and the second identifying what a physical therapist can perceive. The experiment is to find the minimal detectable change in perception and compare this to the minimal clinically important difference.

Asymmetries in different gait parameters (e.g., temporal, spatial, or kinetic) may not be perceived equally. If different parameters are more easily noticed, perception of a specific gait parameter may be used either as an additional metric or as a weighting factor applied to different gait measures. Further, a bias towards perceiving certain measures of asymmetry may impact a physical therapist's perspective of the gait since many clinics rely on observational analysis when prescribing treatments. The misperception of a gait may overemphasize certain attributes of gait that are most noticeable (and hence receive more focus during therapy), but those noticeable traits may not relate most to function. Although there is likely a bias, physical therapists may show less bias towards individual measures of gait asymmetry than untrained individuals. Computer rendered graphics based on different gait patterns may be used to identify the limits of what is perceived as unimpaired when a gait is physically altered, such as asymmetries in spatial and temporal gait parameters or changes in cadence. For example, participants may watch two gait videos side-by-side and choose the gait pattern that appears less impaired. The participants may also be asked to rate the level of gait impairment on each video on a 7-point Likert scale. FIG. 9 shows a version of this experiment used to collect the preliminary data. A web-based design can reduce experimenter bias since little interaction will occur between the participants and the experimenters during the testing sessions. A repository of the gait videos, parameters, and subjective ratings may be made be available on the web and continuously updated.

FIG. 10 shows plots of participants' responses to gait videos (showing a combination of asymmetric and symmetric Step Length and Swing Time features, L=Step Length, T=Swing Time, s=symmetric, a=asymmetric) in categories corresponding to “Very Normal,” “Normal,” and “Very Abnormal” in response to the question: “How normal does this gait appear?” The perception results in FIG. 10 indicate that asymmetries in step timing are more noticeable than similar percentage changes in step lengths. Interestingly is that perception was still rated mostly normal when some parameters were more than 10% asymmetric. Typical rehabilitation would suggest that 98% asymmetry would be the ideal gait pattern, but a balanced gait pattern suggests that anywhere in the 50-90% range could be ideal.

1.4.4. General Data Collection and Analysis Procedures

In various embodiments, kinematic data can be collected using a motion capture system, and a Bertec instrumented force plate mounted under the split-belt treadmill can be used to collect ground reaction forces. Both may be used to calculate the spatiotemporal gait parameters. Gait parameters can then be recorded while subjects walk on the treadmill at their preferred speed. For example, the motion capture system can be calibrated and static & dynamic calibration trials can be collected for each subject. Participants will typically be asked to wear tight-fitting clothing to minimize marker movement. Subject parameters such as body mass, height, stroke (location, type, severity), and dominant side can be recorded. Passive reflective markers may be attached to the subject using a combination of neoprene straps and double-sided adhesive collars.

To assess walking symmetry, the degree of asymmetry (D) in each of the multiple gait measures between the affected, or physically altered, side (A) and the unaffected side (U) can be calculated using: D=(U−A)/[(U+A)/2]. D=0 represents complete gait symmetry, and a positive and negative number represent unaffected side asymmetry and affected side asymmetry, respectively.

1.5 Impacts

While the disclosed notion of balanced and effective gait patterns is a new paradigm that could fundamentally shift the goal for gait rehabilitation, these new desired outcomes could have a substantial impact on the many individuals receiving rehabilitation after a stroke or amputation. Each year, over 795,000 Americans experience a new or recurrent stroke, and more than 6 million individuals have survived a stroke and are living with the debilitating after-effects. Many of the same principles of a balanced gait can also benefit individuals with a lost lower limb. Approximately 1 million Americans have an amputated lower limb. The loss of a lower limb limits many activities of daily living, such as walking and running.

In rehabilitation research, scientists and clinicians attempt to perform well-controlled studies that test one intervention at a time while minimizing differences in other variables. However, in rehabilitation practice, several techniques are often employed concurrently to try to correct a problem. The degree to which different training techniques complement or interfere with each other is unknown. Methods and systems of the present disclosure can help distinguish these interactions, which will help guide clinical practice by either understanding how to combine them or limiting which ones should not be combined.

The literature has continued to show that patients are dissatisfied with their options for training after they are discharged from the rehabilitation hospital/clinic and 85% of patients would prefer a home-based rehabilitation solution. The ability to train at home means individuals can train more often, which leads to better results in motor relearning and can maintain individuals' ability to perform activities of daily living. In accordance with the present disclosure, exemplary systems and methods of the present disclosure can be adapted for use at home. The GEMS demonstrates that motion applied to the foot, height, and weight can easily be adapted to home-use. Pulling on the hip and pulling on the foot, however, are unlikely to be adapted to home-use due to the difficulty of a portable setup to apply force through cables while walking.

In various embodiments, components of exemplary systems and can be integrated within a phone app, such as the therapy aimed at correcting asymmetric gait at home. An exemplary app can be designed to evaluate a person's gait using the built-in gyroscopes and accelerometers, much like existing methods do (e.g., Fitbit, Apple Watch). Based on the step timing, sounds will be played to the person indicating when to place their feet down. The asymmetric timing can be adjusted in real-time to provide a constant level of training, in accordance with various embodiments of the present disclosure.

As discussed, machine learning can be used as a core component in implementing methods and systems of the present disclosure, in various embodiments of the present disclosure. For example, machine learning models, such as convolutional neural network(s), can be trained and tested on perception gait ratings of video subjects from participants and then used to classify new videos of a subject which can be provided as feedback that is applied as part of or in addition to a joint metric with other feedback gait parameters.

2. Examples of Multiple Simultaneous or Coordinated Therapy Interventions Systems and Methods 2.1. General Concepts

This section further discusses systems and methods using two simultaneous perturbations: treadmill training and auditory cueing with symmetric or asymmetric ratios between legs. The two simultaneous interventions are discussed above in connection with Section 1.1.3. The disclosure below shows that combinations of tied-belt treadmill training and symmetric rhythmic auditory cueing (RAC) enhance gait performance more than symmetric RAC overground. In addition, combinations of split-belt treadmill training and asymmetric rhythmic auditory cueing (ARAC) further enhance gait performance.

The treadmill walking enforces adjusted speed to the user. Inadequate walking speed is a major contributing factor for inefficient gait among stroke survivors. When matching speed on a treadmill, the spatiotemporal differences between post-stroke and healthy matched subjects were reduced. While tied-belt treadmill walking (same speed on both sides) is able to improve walking speed (and increase efficiency as a result), it is not effective in improving gait symmetry long-term. Exaggerating asymmetry of an impaired gait (error augmentation) using a split-belt training (SBT) has shown a better post-training aftereffect. This aftereffect can be linked to the compensation mechanism in gait that is applied and stored through proprioceptors (sensory receptors within the muscles) during training. For instance, a person is driving a car with misaligned wheels pulling the car slightly toward the left. The person needs to constantly adjust the steering wheel to compensate for the tilted wheels to stay in a straight line. After a while, if the person drives a new car, the person will notice errors in steering the wheels toward the right when trying to go straight even though your new car is perfectly aligned. This is the effect of error augmentation training, and it is used for augmenting the asymmetry in stroke survivors with the goal of achieving reversed aftereffect (symmetric gait) post-training. While using an SBT has improved gait speed and interleg parameters such as step length, it does not show any aftereffect on intraleg parameters such as stride duration. It is possible that the lack of sensory feedback during training may account for the limited effects. In addition, accessing an SBT is not easy and most patients require visits to rehabilitation facilities for training. Another major challenge of this type of training is transferring the adapted patterns of SBT to overground walking and into the everyday home environment.

Rhythmic Auditory Cueing (RAC) is a training approach for gait symmetry that can affect a different range of gait parameters with a different impact than SBT. There is a strong connection between auditory feedback and the motor control system throughout the central nervous system (CNS). When a subject is walking, RAC influences them by providing musical rhythms to cue motor function. Research has tested the effect of RAC with various features, ranging from isochronic to biologically varied and from metronomes to music-based beats. Previous research has used RAC in which the cues to the left and right are the same duration. A preliminary study evaluates the effectiveness of Asymmetric Rhythmic Auditory Cueing (ARAC) in which the cue duration differs between the left and right legs. The adaptation of each leg has been shown to be independent of the other side. Therefore, it is hypothesized that ARAC is able to train each leg to their cue duration independently. Auditory cueing is more accessible than SBT and can be easily used in any environment (like home) through any sound playing device. However, the effectiveness of auditory cueing varies among patients depending on their rhythmic skills. While some patients have shown significant positive results, others show no or negative results.

Rehabilitation therapies using a single intervention (such as auditory cueing or treadmill training alone) are not capable of creating widespread gait changes that are effective for various patients who demonstrate different aspects of asymmetric gait. The combination of rehabilitation therapies shows better results than a single rehabilitation therapy, giving more ability to control parameter changes and the gait outcome performance as a result. Multiple-rehabilitation therapy includes two or more interventions applied simultaneously with the purpose of enhancing performance by engaging more sensory feedback and increasing the control over adjusting gait parameters. Single rehabilitation therapy can train gait to improve some parameters while making no change or worsening others. A multimodality approach can increase the outcomes of clinical training and focus on the individual's needs and capabilities. Combinations of tied-belt treadmill training and symmetric RAC show to enhance gait performance more than symmetric RAC overground.

The mechanism of gait response from concurrent external stimuli is not completely understood yet. This section discloses the interworking of gait performance under two therapy interventions applied simultaneously: treadmill training and rhythmic cueing. It is hypothesized that the superposition principle applies to gait response; the net gait response caused by two or more stimuli (rehabilitation interventions) is the sum of the gait responses that would have been caused by each stimulus individually. FIG. 11 demonstrates the predicted behavior of the neuromusculoskeletal system under this hypothesis. Baseline rectangle 1102 indicates different gait parameters level of asymmetry. Bars (u) in the upper rectangle 1104 and bars (d) in the lower rectangle 1106 indicate effects from two rehabilitation therapies (Therapy 1 and Therapy 2) applied separately. The rectangle 1108 on the right is the hypothetical behavior of gait under multiple-rehabilitation therapy with therapies 1 and 2 applied simultaneously. Arrows indicate direction of asymmetric effect. Each parameter gets affected differently under different interventions (single rehabilitation therapy). Bars (p) summing arrows for corresponding parameters in the rectangle 1108 on the right indicate estimated gait caused by simultaneous application of therapies 1 and 2. Thus, if multiple-rehabilitation therapy (combining two interventions) is applied, it is estimated that gait combines the response with a linear model.

2.1 Exemplary Methods

Rationale: To investigate the linearity of the gait response, one exemplary experiment evaluates two properties of linear systems: (1) additivity examines if the gait asymmetry under multiple-rehabilitation therapy is the sum of the gait asymmetries under single therapies, and (2) homogeneity evaluates if the proportion of asymmetries applied in each therapy are transferred. Evidence from existing research has given glimpses into the related walking mechanisms. Previous research has not yet studied the relationship between how gait responds under single and multiple-rehabilitation therapy.

The exemplary experiment shows that a combination of SBT and ARAC improves gait asymmetry of a person. Each method creates a type of asymmetry in the gait parameters. It is hypothesized Each of these therapies will affect the resulting gait independently of the other. Individual differences are considered between each subject and provide personalized models for the multiple-rehabilitation therapy that takes these differences into account to allow for future customization.

Experimental design: A prospective cohort study is conducted to test the hypothesis. Each subject completed four different trials. Trials were at least 24 hours apart to make sure the residual effects from the previous trial had washed out. Multiple-rehabilitation therapy is applied by combining treadmill training and auditory cueing (FIG. 12A). FIG. 12A shows types of therapeutic interventions where height of each shape represents relative magnitude 1202 of intervention. FIG. 12B shows the experimental design of each trial. Each trial incorporated one out of the four combinations depicted in FIG. 12B. For both asymmetric interventions, a 2 to 1 ratio is used between left and right. In trials 1 and 2, only one of the therapies has an asymmetric 2:1 pattern 1204 while the other stayed symmetric 1206. In trials 3 and 4, both therapies were applied with an asymmetric ratio of 2:1 in order to test the effect of the error augmentation mechanism. In trial 3, the asymmetric ratios were matched congruently where the faster belt was on the same side of the longer cue. In trial 4, the asymmetric ratios were matched incongruently, where the faster belt was on the same side of the shorter cue.

Participants: 16 healthy subjects (5 females; mean weight=70.5 kg, SD=13.0; mean height=170.9 cm, SD=9.6) with no prior history of gait impairment and no gait injury in the past 12 months form the final group. The subjects are divided into two equal groups of eight. Each completed all four trials (FIG. 12B) in random order. The side of the faster tread was switched in trial 2: left belt was faster for group A and the right belt was faster for group B.

Definitions, data collection, and procedure. Here, the term ‘tied-belt’ is used when the treadmill belts have the same speed, and split-belt' is used when each belt has a different speed. Percent asymmetry is defined as left-right' divided by the sum of left and right. This definition is chosen over ‘slow-fast’ because the side of asymmetry in the trials is considered. 5 time windows are calculated to compare the results of each trial. Each time window is the average of asymmetry between 25 left steps and 25 right steps (50 consecutive steps). Each subject took a few steps to adjust their walking after the start or stop of the training. Therefore, the time windows are located just after the first 10 steps or before the last 10 steps of each phase to make sure gait reaches stability (we first averaged the steps within each subject, then averaged the overall group). All experiments and data recording were conducted using the Computer Assisted Rehabilitation Environment (CAREN) system. CAREN system contains an immersive virtual reality environment, a 3D motion capture system, 6-degree-of-freedom motion base, instrumented dual-belt treadmill, and integrated force-plates. Reflective markers can be used on the lower body joints including metatarsal, heel, and lateral malleolus of the ankle to keep track of lower limb movements. Force-plates under the treadmill can capture ground reaction forces (GRF). Sound cues can be played through surround audio speakers mounted on the ceiling of the safety cage. Each cue can be the recorded standard dictionary (American English) pronunciation of the word ‘right’ or ‘left’. Subjects were instructed to adjust each foot landing with their respective cue. Data were recorded at 100 Hz frequency. Prior to the first trial, subjects practiced on the tied-belt treadmill until they felt comfortable (approximately 3 minutes). During this time, they were asked to choose their comfortable speed while the operator changed the speed of the treadmill by increments of 0.1 m/s. Then the step time of the tied-belt walking was calculated by averaging 10 consecutive steps on the treadmill with the set comfortable speed. The fast belt was calculated by 4/3 of the comfortable speed, and the slow belt was half of the fast belt (⅔ of the comfortable speed) to keep the same average speed with a 2:1 ratio. The same process was implemented for the duration of asymmetric cues in ARAC. The cues are spaced isochronously based on the step time. For instance, if the average step time of a subject is 750 ms, a 2:1 ratio of ARAC has a 1000 ms cue on one side and a 500 ms cue on the other. Each trial took 23 minutes to complete (FIG. 12C) and had three phases: 3 minutes of baseline (tied-belt and no sound), 15 minutes of adaptation using one of the trials 1 through 4 combinations, and 5 minutes of post adaptation (tied-belt and no sound). FIG. 12C shows a duration of each section in the experiments. Rectangular shapes 1208, 1210, 1212, 1214, 1216 show the approximate location of time windows used for analysis. Each window includes 25 left and 25 right steps. The time windows are located after the first 10 steps or before the last 10 steps of each phase. Five time windows (BL baseline 1208, EA early adaptation 1210, LA late adaptation 1212, EP early post-adaptation 1214, and LP late post-adaptation 1216) are measured in the experiment.

Data analysis: This experiment is a within-subject design. Multiple linear regression analysis can be conducted to test the hypothesis using two explanatory variables (trials 1 and 2) to estimate the outcome of two dependent variables (trials 3 and 4). Statistical analysis can be performed using IBM SPSS Statistics 26. The regression analysis reported the r-squared values, the significance levels using p value (0.05), and the collinearity diagnostics of the linear model. Each coefficient of the model was calculated based on 30 data points (2 trials 3 gait parameters 5 time windows).

2.2. Results

Gait parameters: Subjects have an average comfortable speed of 0.91 m/s (SD=0.21) with average step time of 0.63 s (SD=0.093). Spatiotemporal parameters, as well as GRF, can be calculated. FIG. 13 indicates the average of percent asymmetry for step length, step time, and peak vertical force for group A during the three phases of the experiments (baseline, adaptation, post-adaptation). FIG. 13A shows trial 1 (tied-belt+ARAC) has a larger effect on step time compared to step length asymmetry. Auditory sensory feedback shows to influence largely the temporal parameters in the gait. Trial 2 (split-belt+RAC) in FIG. 13B affects both step length and step time (step length slightly more than step time). This result shows that SBT engages interlimb parameters. Trials 3 (congruent) and 4 (incongruent) use both SBT and ARAC effects at the same time. FIG. 13C with the congruent combination is indicating a sum effect of trials 1 (tied-belt+ARAC) and 2 (split-belt+RAC), while FIG. 13D with the incongruent combination is indicating the difference of trial 1 (tied-belt+ARAC) minus trial 2 (split-belt+RAC). In all SBT trials, there is an opposite change in asymmetry during early post-adaptation compared to early adaptation, which indicates the neural system has temporarily stored the applied asymmetry6. Group B confirms the similar behavior of combined effect and storage of adaptation through overcorrection mechanism.

2.3. Linear Model

The superposition hypothesis proposes that the resultant change in gait under two stimuli is the linear combination of changes under each stimulus applied individually. To test the applicability of this hypothesis for gait response under asymmetric interventions, a linear model is developed for the three measurements (step length, step time, and peak vertical force) indicated in FIGS. 13A-13D. Two asymmetric stimuli, SBT and ARAC, are applied both individually and combined. Trial 1 and 2 include only one asymmetric stimulus, while trials 3 and 4 apply both stimuli asymmetrically. A model that estimates the gait response for the two later trials can be found based on the first two trials' gait responses. Equation 1 shows the linear model equations:


Trial3estimate=C1×Trial1+C2×Trial2


Trial4estimate=C1×Trial1−C2×Trial2   [Equation 1]

C1 and C2 are constant coefficients of the model and stay the same in both equations because the interventions are applied with the same ratio in trial 3 and 4. The direction of applied asymmetry between trial 3 and 4 is changed to create the congruent and incongruent effect. The negative sign in trial 4 estimation indicates this difference. The two coefficients can be calculated by minimizing the root mean square error of the models at the same time. Trial 3 is the congruent combination because both directions of asymmetry in ARAC and SBT guide the gait toward the same asymmetric side. Trial 4 is the incongruent combination since the asymmetric direction of SBT and ARAC guide the gait toward opposite sides of asymmetry. The result of the fitted model is compared to the actual data (FIGS. 14A and 14B) for both groups A and B. This linear model estimates the behavior of three gait parameters in spatial, temporal, and kinetic areas over the three phases of the experiments for two combinations of asymmetric interventions.

[C1, C2]values are [0.84, 0.89] for group A and [0.61, 0.90] for group B. The overall r-squared values for group A and B models are 0.96 and 0.95, respectively. The r-squared values indicate the linear model explains more than 95% of variation within the data. The statistical test also calculates the p values of variables in the regression model. The p-values of both independent variables (trial 1 and trial 2) are 8.4e-20 in group A and 2.6e-19 in group B, indicating very strong evidence in favor of the hypothesis. The results show that combined asymmetric changes under trial 1 and 2 are significantly associated with changes in the response of trial 3 and 4. No collinearity is found among trials 1 and 2, having values of the variance inflation factor of 1 for all coefficients. While C2 values in both groups are close (0.89 and 0.90), the C1 value for group B is 0.61, indicating less effect of ARAC in group B compared to group A. After close examination, two subjects in group B showed close to zero or negative gait response under trial 1 (tied-belt+ARAC). Running the model for group B without the two outlier subjects gives values closer to group A: C1=0.77 and C2=1.00.

Personalized coefficients: The disclosed linear model of the gait response applies the superposition principle to asymmetric gait behavior under two simultaneous stimuli for the averaged results of all subjects. While the model presents a great fit for the averaged data supporting the hypothesis, it does not provide the optimal model for each individual. Thus, personalized coefficients are calculated for each subject by minimizing the root mean square error of the linear model (Equation. 1) for each person. FIG. 15 indicates the whisker plot of the coefficient values for the optimal personalized linear models. Two subjects in group B have close to zero or negative C1 values for their best fit models. Training based on the auditory sensory feedback have shown none or even negative responses from participants depending on their rhythmic synchronization capabilities. As a result, the first quartile of C1 in group B was stretched to near zero.

2.4 Discussion

Reciprocal movements such as walking contain three stages: initiation, cyclic pattern, and termination of the movement. CNS involvement has been suggested to vary between these stages. While initiation and termination are regulated at the supraspinal level, cyclic patterns are generated at the spinal level by central pattern generators (CPGs) and motor neurons. Therefore, the involvement of the supraspinal level in pattern generation is minimal and indirect. While there is strong direct evidence for the existence of CPGs in other vertebrates such as cats, there is indirect yet strong evidence supporting the existence of CPGs in humans. Moreover, research has shown that the brain does not control every detail of every movement. Instead, it only controls the endpoint, in this case, the final placement of the foot. It combines groups of muscles into modules for controlling tasks such as balance or walking. Modules simplify control of the coordination of movement by the nervous system and create flexibility within the muscles, sensory feedback, and neurons. In patients suffering from stroke, damage to the brain causes these modules to integrate together in order to make them easier to control. Therefore, the default walking module changes. The default walking module is the preferred walking style when no intervention is applied, whether a person has a healthy or an impaired gait. Now, due to the interconnections of the neuromusculoskeletal system, damage to the brain does not mean that the ability to walk normally and create new modules is completely lost. Cerebellar damage limits gait adaptation, but the capability for rehabilitation and gait retraining is retained after damage to the cerebral cortex. Therefore, stroke survivors with cerebral damage are capable of (re)learning symmetric walking patterns. (Re)training the muscles and restoring the pathways of neural control of movement through rehabilitation therapy have shown promising results. However, single rehabilitation therapy comes up short of complete gait restoration.

However, this disclosure shows that multiple rehabilitation therapies with different doses can improve gait restoration better than a single rehabilitation therapy. For example, two interventions are applied with 1:1 or 2:1 ratios at the same time in four different trials, measuring gait response each time. By combining two rehabilitation therapies, the superposition principle applies to the gait response under combined treadmill training and rhythmic cueing. Using the ARAC technique, an innovative method developed for the first time to match 2:1 asymmetry in split-belt with the same asymmetry in sound ratios. Auditory cueing (RAC or ARAC) has a feedforward mechanism where gait parameters adjust in anticipation of the upcoming cue. In this rehabilitation therapy, the ability to minimize the error between the predicted cue and the actual cue plays an essential role. Treadmill training (split-belt or tied-belt) has a feedback mechanism where the subject reacts to the change of speed or asymmetry, activating interleg control, which adjusts the gait parameters to bring the pattern back to the default walking pattern. While the treadmill trains gait by engaging muscles and proprioceptors into a different walking pattern, auditory cueing requires conscious attention to auditory sensory feedback.

Modeling gait response can lead to a better understanding of the neuromusculoskeletal system performance under multiple stimuli. A gait response model can also help with developing personalized multiple-rehabilitation therapies that encompass multitudinous aspects of individual capabilities. In this disclosure, a linear model is proposed for the gait response under simultaneous stimuli (interventions) and calculated the percentage of contribution of each stimulus to the final gait asymmetry in each subject. The coefficient values indicate the relative effectiveness of each intervention in the combined trials. A value of one for C1 or C2 indicates the full effect is realized. A coefficient between zero and one would indicate the effect of the intervention was partially transferred. C1 and C2 values for both groups are higher than 0.6 and less than 1, which indicates that less than 40% of the therapeutic effects of each intervention is lost when the therapies applied simultaneously. Three out of the four coefficients are higher than 0.84, meaning the loss of therapeutic effects of their corresponding interventions are 16% or less. For example, C1=0.85 and C2=0.91 for group A means that 85% of the ARAC effect and 91% of the SBT effect were demonstrated across step time, step length, and peak vertical force during combinations of ARAC and SBT. The high r-squared values and the level of significance confirmed that the suggested model estimates the gait response under two simultaneous stimuli with high accuracy.

While this model shows the average response of all subjects, it might not represent the best estimate for each subject individually. Rhythmic capability, muscle strength, physique, and other personal aspects can affect the gait response of individuals under asymmetric interventions. Level of impairment also change the effectiveness of different rehabilitation therapies between individuals. For this reason, it is important to look at the variability of the model among individuals. It is vital for the improvement of personalized therapies that individual characteristics, especially individual gait response to various inputs from different therapies, to be taken into consideration when prescribing a multiple-rehabilitation therapy.

Calculating personalized coefficients from gait responses of 16 healthy subjects showed more than 70% of the coefficient values are between 0.54 and 1, meaning that the majority of the healthy subjects demonstrated more than 50% of each individual intervention when combined. A coefficient of more than 1 for either C1 or C2 would mean adding an intervention has acted as a catalyst and emphasized the corresponding therapy method. A coefficient of zero (or close to zero) for either C1 or C2 would mean that the subject was not able to respond to the corresponding intervention or the effect of the corresponding intervention was lost completely or overlapped with the other intervention during the combined ARAC and SBT trials. Five subjects had a coefficient ranging between 1.06 and 1.34. This could indicate that the addition of a therapy method has made the other method more impactful. Two subjects indicated close to zero or negative coefficients during trial 1 (ARAC+tied-belt). This is not an unexpected outcome in therapies that are based on rhythmic auditory feedback. Previous research has shown that while some people have a positive response to rhythmic cueing, others might have none or even negative response. The performance of rhythmic cueing has been connected to the rhythmic skills of people.

If a minimum of 40% is considered as a considerable demonstration of asymmetries in gait response, more than 90% of the subjects are considerably affected from each intervention during both combined trials. While the exact coefficients for each subject might vary depending on their strengths and weaknesses, both the fitted linear model on the averaged data (FIGS. 14A and 14B) and the close range of the whisker plots for individuals (FIG. 15) indicate that neuromusculoskeletal system can linearly combine the effects of two simultaneous rehabilitation stimuli on gait asymmetric response. Experimental results showed that the additivity principle was met; however, the coefficients not being one indicates that the homogeneity was not fully applicable. Therefore, the superposition principle was largely applied. This result can lead to a better and improved combination of therapies that accommodate the needs of patients as well as leveraging their strengths for better outcomes.

This disclosure may have the potential to significantly impact gait rehabilitation in people with neuro-logical disorders such as stroke. Additional understanding is needed to get to this level, such as the development of accurate, reliable, and practical assessment methods to determine to what extent a patient responds to the individual or jointly applied sensory stimuli. Once the methods are ready to use, therapists can easily identify and apply optimal levels of sensory stimuli needed for personalized gait rehabilitation based on the patient's capabilities and needs. For example, a patient, who experiences significantly altered step time but minimally impaired step length after stroke, may require a great level of auditory cue and a minimal level of visual cue to maximize gait recovery. The linear regression model introduced in this study can help develop the methods that effectively assess the patient's responses to the sensory stimuli.

This disclosure achieved two discoveries in the path of understanding gait response. First, ARAC is able to adapt different sides of gait by applying asymmetric cues with a 2:1 ratio, meaning that auditory sensory feedback can independently access each side of lower limbs. Second, gait response for these gait parameters can be mainly modeled as a linear system with the possibility of quantifying the contribution of various stimuli at the same time.

3. Examples of Real-Time Feedback Control of Split-Belt Ratio Systems and Methods 3.1. General Concepts

This section further discusses systems and methods using split-belt treadmill training with real-time feedback. The split-belt treadmill training with real-time feedback is discussed above in connection with Section 1.1.3. The disclosure below shows that split-belt treadmill training with real-time feedback enhances gait performance.

Split-belt treadmill training is a common gait therapy used to modify step length asymmetry by adjusting left and right tread speeds individually. This therapy's primary goal is to assist with gait rehabilitation in patients who have had a stroke. However, current split-belt approaches pay little attention to the individuality of patients by applying set tread speed ratios (e.g., 2:1 or 3:1). This generalization results in unpredictable step length adjustment. To customize the therapy, this section explores the rehabilitation capabilities of a live feedback system that modulates split-belt tread speeds based on real-time step length asymmetry. Materials and Methods: A proportional-integral (PI) controller is used to maintain a specified step-length asymmetry in 14 healthy volunteers. Subjects participate in two 1.5-hour sessions scheduled approximately one week apart, during which they were asked to walk on the CAREN split-belt treadmill system with a boot on one foot to impose asymmetrical gait patterns. The subjects walk under the following conditions: 3-minute baseline, 10-minute baseline with boot applied, 15-minute training (boot-baseline asymmetry exaggeration by 6% in the first session and personalized in the second) and after-effect (return from training to tied-belt) evaluation, and 3-minute boot removed. After the first session, a linear model between baseline asymmetry exaggeration and after-effect improvement was utilized to develop a relationship between target exaggeration and target after-effect. In the second session, this model predicts a necessary target asymmetry exaggeration to replace the original 6%. This prediction is intended to result in a highly symmetric after-effect. Results and Discussion: 78.6% of subjects can develop a successful relationship between asymmetry exaggeration and decreased asymmetry in after-effect. 63.6% of subjects in this successful correlation group can have second session after-effects of <4% asymmetry. Conclusions: The use of a PI controller to modulate split-belt tread speeds demonstrates itself to be a viable method for individualizing split-belt therapy. Further, this finding paves the path towards more effective control of other commonly used gait interventions using real-time feedback controller algorithms.

3.2. Introduction

Conventional split-belt therapy techniques do not consider the individuality of participants. This is due to the use of set tread speed ratios, which necessarily assume uniform response from the participant population. As participants respond to training at different rates, individualization of therapy is a strategy that must be further developed. Therefore, this section investigates the use of a real-time feedback controller algorithm focused on maintaining a set step length asymmetry. The purpose is to evaluate the controller's ability to achieve target asymmetries in participant gait. Additionally, we determine the validity of an algorithm that predicts intervention outcomes for the purpose of individualizing therapy to achieve after-effect step length asymmetry values <4%.

3.2.1. Split-Belt Therapy

Split-belt treadmill training is a common approach taken to modify gait patterns, particularly step length asymmetries. The belts are set such that one is “fast” and the other is “slow.” The common result of this perturbation is that the step length on the fast belt will be longer while the step length on the slow belt will be shorter. Upon a sudden return to a tied-belt state, the participant expresses more symmetric step lengths, opposite in direction of those induced by the split-belt intervention. The after-effects imposed by various split-belt ratios between 1:1 and 1:3 can be used. These experiments demonstrated a correlation between the speed ratio of split-belt intervention and the severity of step length asymmetry observed in adaptation and washout periods. In the case of post-stroke participants, the transfer of after-effects from split-belt therapies from treadmill to ground shows that the “reversed after-effect” imposed by a 2:1 split-belt ratio in both healthy and post-stroke participants can be still partially present after transfer from the treadmill to over ground. In fact, this section suggests that post-stroke patients retain split-belt induced after-effects significantly better than healthy participants, indicating that treatments applied to healthy participants can potentially be expected to yield better results when applied to post-stroke participants.

The split-belt approach can also be modified in an effort to determine whether sudden or gradual deviation from tied-belt to split-belt has any effect on participant performance. In an experiment, 8 participants were brought from 0.7 m/s tied-belt to 2:1 split-belt (1.4 m/s to 0.7 m/s) by an acceleration profile of 0.02 m/s2 every 20 strides and a separate 8 participants by an acceleration profile of 10 m/s2. This experiment can reveal a novel kinetic pattern at the hip joint of those trained with the gradual split belts. Specifically, the 10 m/s2 acceleration scenario resulted in a decrease in work at the slow hip joint. Conversely, the slower acceleration scenario resulted in little to no difference in work done between the fast and slow hip joint. This experiment indicates that there is a difference in adaptation when applied with gradual vs. sudden acceleration. Therefore, the overall effects of their therapy need to be further explored to determine its efficacy when compared to standard sudden split-belt training.

3.2.2. Inducing Asymmetry

As recruitment of post-stroke participants is significantly more difficult than recruitment of healthy participants, the ability to impart gait asymmetries to healthy participants is vital. Commonly used strategies for inducing gait impairments are the application of weight, height, or both to participants. Applying a weight of 5% participant body mass is known to induce step length asymmetries as high as 7%. Further, inducing a leg length discrepancy of only 3 cm has produced step length asymmetries of approximately 6%. Previous research has also applied various combinations of weight and height to participants' legs to induce asymmetries. A study found that there was no significant interaction between the effects produced by the addition of weight and height. Essentially, height and weight perturb gait independently, and their modifications add up when applied at the same time. Analysis of these results allows for determination of which intervention is more efficient. The addition of a 5.2 cm leg length discrepancy induced approximately a 6% asymmetry, while application of no leg length and a 4.6 kg weight only induced approximately a 2% asymmetry. Thus, to create a large step length asymmetry without adding so much weight that the participant becomes exhausted, addition of leg length is beneficial.

Initially worsening gait asymmetries causes the central nervous system to adapt such that the after-effects improve symmetry when the perturbation is removed. However, there has been no investigation into an intervention that is able to predict the after-effects within participants. To individualize split-belt therapy and potentially predict gait parameters, it is necessary to control specific gait parameters in real-time. This section implements personalized split-belt speed ratios to improve control of gait parameters and accomplish symmetric walking.

3.3. Exemplary Methods 3.3.1. Participant Selection

Eligibility of volunteers is determined based on stamina and lower-limb injury, as well as being 18 years of age or older. Participants were only deemed eligible if they are free of lower limb injury for a minimum of 6 months prior. Further, volunteers were told that they would need to walk for approximately 35 minutes on a treadmill with a weight added to one leg for most of the session. They were only scheduled if they are confident in their abilities to complete this task on two separate occasions. Upon arrival, participants were asked to fill out a survey to record their age, sex, height, weight, and activity level.

3.3.2. Data Collection

The real-time feedback experiments can be conducted on a CAREN (Computer Assisted Rehabilitation Environment) system. In some examples, the CAREN system is the Motek CAREN system. However, it could be any other suitable system having a processor and a memory. A suitable number of cameras controlled by motion capture software can be used. For example, a total of 10 Vicon cameras controlled by Nexus motion capture software can be used. These cameras tracked 11 markers on each participant. However, the number of markers is not limited to 11 markers but could be any suitable number of markers for tracking motions of a participant. The split-belt treadmill capabilities of the CAREN system may be controlled by a script written in a suitable software. In some examples, the software may be the D-Flow control software. A real-time sorting algorithm was written in D-Flow to label and track the 11 markers. Offline analysis verified the accuracy of the measured parameters. In the first moment of the trial, prior to belt movement, markers were labeled based on the assumption that the participant was standing upright and still. Distances are calculated between every marker, and these distances, alongside standard gait assumptions, are used throughout the experiments to keep the markers appropriately sorted in real-time. Step lengths are calculated in real time by calculating the distance between heel markers for every step taken. A step is determined to be when the force plates received a reading of >200N (>150N in the case of one small participant). The marker motion in Vicon is closely watched to verify the real-time calculations, ensuring that step length asymmetries are measured correctly.

Step length is measured in real-time for all trials conducted and used as the driving metric for split-belt adjustment in the feedback controller. Step length is defined as the distance between heel locations upon heel-strike. The definition is used for the real-time algorithm. Thus, as volunteers walk, step length is recorded upon each heel strike as heel Z coordinate location minus contralateral heel Z coordinate location. Many other parameters are gathered in real time as well, such as joint angles and heel-strike/toe-off forces, but this section uses real-time monitoring of step lengths to influence step length asymmetry.

3.3.3. Feedback Controller

FIG. 16 shows a block diagram of real-time feedback algorithm 1600 controlled and executed by a real-time feedback controller. The feedback controller algorithm 1600 shown in FIG. 16 can enforce a specific step length asymmetry in participant gait. Therefore, at 1602, the researcher conducting the experiment would input a desired LL% asymmetry (definition of LL% seen in Equation 2). The LL% metric is used to measure both the magnitude of asymmetry as well as the direction, as positive values indicate a longer left step length and negative values indicate a longer right step length. Every 20 seconds, the left and right step lengths are averaged at 1604, and this average LL% is compared to the desired LL% at 1606. At 1608, the error present is then sent to the PI controller 1610, which utilizes proportional and integral gains to calculate a necessary adjustment for each tread at 1612 while keeping the average speed between the belts constant. This adjustment was processed by a limiter 1614 that only allows 0.15 m/s to be added to/subtracted from each belt per adjustment. The limit for a speed adjustment can be any other suitably predetermined limit (m/s). Further, this limiter 1614 enforces a maximum ratio of 3:1. The limit for a speed ratio can be any other suitably predetermined ratio. Thus, at 1616, the limiter 1614 sends the processed adjustment to the treadmill 1618, which then modulated the left and right belt speeds at 1620 and 1622. At 1624, the participant's step lengths are measured as they walked on this new ratio, and the process continues. Equation 2 defines left-long (LL%) step length asymmetry below.

LL ⁢ ⁢ % = L ⁢ S ⁢ L - R ⁢ S ⁢ L A ⁢ S ⁢ L * 1 ⁢ 0 ⁢ 0 ⁢ % ⁢ ⁢ LL ⁢ ⁢ % = Left ⁢ ⁢ Leg ⁢ ⁢ Long ⁢ ⁢ Percentage ⁢ ⁢ L ⁢ ⁢ S ⁢ ⁢ L = Left ⁢ ⁢ Step ⁢ ⁢ Length ⁢ ⁢ R ⁢ ⁢ S ⁢ ⁢ L = Right ⁢ ⁢ Step ⁢ ⁢ Length ⁢ ⁢ A ⁢ ⁢ S ⁢ ⁢ L = Average ⁢ ⁢ Step ⁢ ⁢ Length [ Equation ⁢ ⁢ 2 ]

Participants were instructed to walk on the tied-belt treadmill with an initial speed of 0.8 m/s. They were then asked to tell the investigator to speed up or slow down the treads until they felt comfortable. This speed is reported as the participant's “comfortable walking speed.” The 11 markers 1802 were then placed on the participant according to FIG. 18. The protocol is outlined in table format in FIG. 17. At 1702, participants walked for three minutes at tied-belt comfortable speed to gather an overall baseline. During this period, investigators were able to see the asymmetries, calculated using Equation 2 every 20 seconds, but they were only saved and not used to adjust the tread ratios. After the baseline trial, at 1704, a boot 1804 that added approximately 4.5 cm and weighted approximately 1 kg was added to one of the participant's legs. It was added to the left foot if the participant was right-handed and to the right if left-handed. It should be appreciated that the weight and height of the boot can be suitably adjusted. They were then asked to walk for 10 minutes under the same tied-belt condition. The last nine 20 second step length measures (approximately the last 3 minutes) were averaged and considered the boot-baseline step length asymmetry. This value was then exaggerated by 6% (i.e., moved further from zero) and entered as the target into the split-belt control software. At 1706, participants then walked for 10 minutes with the treads adjusting every 20 seconds according to the software to induce the desired target. Upon completion of the 10-minute training session, at 1708, the treads were returned to a tied-belt session without warning, and the participant's after-effect was observed over a 5-minute “cooldown” period. At 1710, the boot was then removed from the participant's leg, and they were asked to walk for another 3 minutes with tied-belts.

The LL% (see Equation 2) present during minute 3 of the cooldown period is considered the usable after-effect. A linear relationship is then established between 6% exaggeration and the level of improvement described by LL% (bootBaseline)-LL% (afterEffect) for session one. The participant then returned for session 2, where the only difference from the previous session is the target asymmetry. After determining the boot-baseline of the second session's 10-minute trial, the value was entered into the linear relationship previously established. The software then output a predicted asymmetry exaggeration that theoretically results in a perfectly symmetric after-effect. FIG. 19 demonstrates how this relationship is used. While the target asymmetry during the sessions is predicted by analyzing the relationship between baseline asymmetry exaggeration and after-effect improvement over baseline, FIG. 19 relates actual asymmetry and actual after-effect (instead of exaggeration and improvement) to better visualize the linear theory. FIG. 19 shows that the slope of the line developed in session one (considered the evaluation session) can be used to predict a necessary target asymmetry for session two regardless of the participant's baseline.

3.3.4. Data Analysis

Participants can be classified into one of two categories based on the results of the first session: successful correlation 1702 and unsuccessful correlation 1704 (see first branch point of FIG. 20). Successful correlation 1702 refers to the magnitude of the after-effect LL% being less than the magnitude of the boot-baseline LL%. Unsuccessful correlation 1704, on the other hand, refers to the magnitude of the after-effect LL% being greater than the magnitude of the boot-baseline LL%. It is well-known that using a split-belt intervention to exaggerate an existing asymmetry results in an after-effect of decreased asymmetry. Therefore, it is assumed that exaggerating the participant's inherent boot-baseline asymmetry by 6% would result in an after-effect LL% of lower magnitude. However, this is not always the case, and participants who experienced larger after-effects are considered part of the unsuccessful correlation group. There are two subcategories for participants that had an unsuccessful correlation (see right side of FIG. 5). Participants would either undergo a controlled intervention 1706, where the predictor model suggests an attainable exaggeration, or a maximum intervention 1708, where the predictor model suggests an unreasonably large exaggeration that results in a 3:1 ratio within the first five minutes. 3:1 is chosen as a maximum because a greater intervention was deemed unsafe. This categorization criteria is outlined in Equation 3. Equation 3 can indicate decision criteria for classification of maximum or controlled intervention.

⁢ [ Equation ⁢ ⁢ 3 ] L ⁢ B ⁢ S R ⁢ B ⁢ S > 3 ⁢ ⁢ or R ⁢ B ⁢ S L ⁢ B ⁢ S > 3 ⁢ ⁢ within ⁢ ⁢ first ⁢ ⁢ 5 ⁢ ⁢ minutess → Maximum ⁢ ⁢ Intervention L ⁢ B ⁢ S R ⁢ B ⁢ S < 3 ⁢ ⁢ and R ⁢ B ⁢ S L ⁢ B ⁢ S < 3 ⁢ ⁢ for ⁢ ⁢ first ⁢ ⁢ 5 ⁢ ⁢ minutes → Controlled ⁢ ⁢ Intervention LBS = Left ⁢ ⁢ Belt ⁢ ⁢ Speed RBS = Right ⁢ ⁢ Belt ⁢ ⁢ Speed

In either case, participants may have their initial asymmetry worsened. This is because the algorithm suggests decreasing the magnitude of their asymmetry or sending the participant to an LL% opposite in sign of their boot-baseline. This erroneous prediction is a direct result of the unsuccessful correlation previously mentioned. Therefore, participants in the unsuccessful correlation category 2004 undergo either controlled or maximum intervention 2006, 2008, both of which should result in an imposed after-effect asymmetry 2010 (see bottom right of FIG. 20). Given that the feedback controller was unable to benefit these individuals because of the erroneous prediction, only the successful correlation category 2002 will be considered when evaluating the efficacy of the controlled intervention 2014.

There are two subcategories for successful correlations 2002, though the outcomes of these subcategories become more complicated (see left side of FIG. 20). Participants are placed in the controlled intervention subcategory 2014 or the maximum intervention subcategory 2012 by the same criteria seen in Equation 3. In theory, both subcategories 2012, 2014 have the potential to either impose symmetry or produce no effect entirely. Participants are placed in the symmetry-imposed 2018 or the unaffected subcategories 2016 after the second session dependent upon the following criteria (Equation 4). Equation 4 indicates decision criteria for classification of symmetry imposed or unaffected by therapy.


|LL%(AfterEffect)|<|LL%(BootBaseline)|→Symmetry Imposed 2018


|LL%(AfterEffect)|≈|LL%(BootBaseline)|→Unaffected 2016   [Equation 4]

Further, these participants are organized into groups of sufficient 2022 and insufficient symmetry 2020 (bottom left of FIG. 20). It has been previously suggested that gait parameters vary by approximately 4% in normal human gait. While those figures are generally regarding forces in human gait rather than step lengths specifically, it is also known that abnormalities in gait are visible at a point somewhere between 5% and 13% asymmetry. Therefore, sufficient symmetry 2022 is defined as 4%, well beneath the noticeable value. Thus, participants were further categorized according to Equation 5. Equation 5 indicates decision criteria for classification after unaffected symmetry-imposed branch.


|LL%(AfterEffect)|≤4% →Sufficient Symmetry Imposed


|LL%(AfterEffect)|>4% →Insufficient Symmetry Imposed   [Equation 5]

3.4. Results

After the first session, 11/14 participants fall into the successful correlation category 2002. All participants in the unsuccessful correlation category 2004 had extremely similar outcomes in their first sessions. In every case, the split-belt speeds varied a negligible amount. Realistically, these participants could be considered to have undergone no intervention whatsoever. Therefore, the successful correlation participant group 2002 will be considered the entire data set for percentages stated in this section. 63.6% underwent a controlled intervention 2014 (see FIG. 21), while 36.4% underwent maximum intervention 2012 (see FIG. 22). All except one of participants (90.9%) had symmetry imposed 2018. The single participant that did not have symmetry imposed was unaffected 2016 because their boot-baseline was close to zero (−0.11%), and thus the target asymmetry was essentially 0%. 63.6% of participants achieved sufficient symmetry 2022, resulting in a reasonably high success rate for successfully correlated participants after only one training session. In FIGS. 21 and 22, the sufficient symmetry borders in step length asymmetry graphs at the bottom may indicate ±4% of asymmetry.

85.7% of the participants that underwent controlled intervention 2014 achieved sufficient symmetry 2022. On the other hand, only 25% participants that underwent maximum intervention 2012 achieved sufficient symmetry 2022. Thus, the controlled intervention 2014 specifically yielded a very high success rate, while the maximum intervention was not consistently capable of inducing symmetry beneath 4% in magnitude.

3.5. Discussion

FIG. 23 demonstrates the differences between session one training period belt-speeds of participants in the various categories. For participants who fell into the unsuccessful correlation category 2302, the speeds hardly changed during session one training when compared to participants in the controlled intervention category 2306 of the successful correlation. Thus, the return to tied-belt resulted in no significant after-effect. In the case of participants in the maximum intervention category 2304 of the successful correlation, the belt speeds also hardly changed, though often more so than the unsuccessful correlation category. Currently, it is theorized that this issue arises out of an incorrect boot-baseline value. The participants in this category required little intervention to maintain the desired asymmetry, which was supposedly 6% greater than their baseline. However, the lack of required intervention leads one to believe that their true baseline was more asymmetric than what was recorded. Thus, the importance of a true boot-baseline asymmetry is vital, and more research must be done to determine what constitutes “true boot-baseline asymmetry.”

3.5.1. Categorized Outcomes

There exist two groups 2012, 2014 of participants within the successful correlation category 2002: those requiring a controlled intervention 2014, and those requiring a maximum intervention 2012. In some cases, participants displayed a weak correlation, meaning that a 6% exaggeration of asymmetry resulted in a very small improvement in after-effect over baseline. These scenarios were often obvious, as the predictor model would suggest enormous exaggerations in the hundreds or thousands. Clearly these values are unreachable, and so a maximum intervention is necessary. The split-belt was capped at a 3:1 ratio, and these participants reached this ratio within the first five minutes of training, often within the first minute. This allows for a direct comparison of a traditional split-belt intervention with the new controlled intervention. In 75% of the maximum intervention cases, sufficient symmetry was not reached. However, only 14.3% of controlled interventions resulted in insufficient symmetry. This study demonstrates that the real-time feedback controller is highly effective when a successful correlation 2002 exists and the controller does not reach its maximum value. More research should be done to understand why some participants fall into the unsuccessful correlation and maximum intervention categories. Reducing these occurrences would benefit more patients and provide a better understanding of their adaptative limitations. Given that participants who underwent 3:1 intervention exhibited some improvement in step length symmetry, it can be assumed that the participants were only capable of this limited level of improvement. This implies that successive sessions over several weeks would improve their boot-baseline asymmetry until the controlled intervention could introduce more precise rehabilitation. This theory can be applied to a post-stroke case, as stroke participants are able to retain trained gait patterns more effectively.

One participant from the controlled intervention category 2014 achieved insufficient symmetry 2020 (baseline: −15.9%, after-effect: −6.8%), and one participant from the maximum intervention category 2012 achieved sufficient symmetry 2022 (baseline: 12.6%, after-effect: 0.7%). While it was expected that the controlled intervention 2014 would not have a perfect success rate, the outcome of the maximum intervention outlier 2008 was surprising. The precision of the after-effect in this case indicates that this participant was in fact capable of fully symmetric restoration, even though the baseline gathered from session 1 indicated otherwise.

3.5.2. Predictive Model for Session Two

Evaluation of each participant's first session after witnessing the outcome of their second yields a clear pattern. The final categorization of each participant can be predicted with reasonable accuracy using the outcome of the first session. After session one, participants were characterized by the slope of the line relating asymmetry exaggeration and after-effect improvement. For example, if the participant responded to the 6% exaggeration by displaying an after-effect asymmetry 3% less than their baseline, the slope of their line would be 0.5. Thus, negative slopes imply that asymmetry exaggeration worsens the gait of the participant, which does not follow present findings. Therefore, participants can be sorted into successful or unsuccessful correlation groups 2002 or 2004 immediately by the sign of the slope. Further, the magnitude of the slope is useful for predicting the outcome of session two. It was found that very small slopes correspond to very weak correlations, which often resulted in maximum interventions 2012 or 2008. On the other hand, larger slopes corresponded to stronger correlations, meaning participants had a higher likelihood of remaining within the controlled intervention criteria 2006 or 2014. The value 0.25 was chosen because it was half of the smallest slope found amongst the controlled intervention 2014 → the sufficient symmetry imposed group 2022.

This model was tested against the full study population and found to have an 85.7% success rate. The model failed for participants 7 and 14. Participant 7 was the only person to maintain controlled intervention and not achieve sufficient symmetry, while participant 14 was the only person to undergo maximum intervention and achieve sufficient symmetry. Thus, this model can identify outliers with proper adjustment to the cutoff slope value of 0.25 (the true value requires a larger data set for reliable prediction).

Further, this model shows great promise for use in a clinical setting. The purpose of session one was to evaluate participant response to intervention, while the purpose of session two was to train the participant accordingly. The ability to predict a patient's response to the training session allows for informed decisions to be made upon completion of the evaluation session. Not all participants respond to all forms of gait intervention, as has been demonstrated in rhythmic cueing experiments for step time rehabilitation. Therefore, this predictive model gives researchers the ability to identify potential split-belt therapy non-responders almost immediately. If after the first session it is likely that the participant will undergo a maximum intervention (small slope observed), it can be assumed that the individual either requires a larger exaggeration or can be classified as a non-responder. Thus, a second session of greater exaggeration and subsequent use of the predictive model would allow for classification of the patient. This approach increases the efficiency of the therapy by flagging non-responders early in the rehabilitation process so that a more suitable therapy may be suggested.

3.5.3. Controller Accuracy

Thus far, only baseline, target, and after-effect metrics have been discussed. However, an important aspect of the real-time feedback intervention is the achieved asymmetry. The “real-time effect” was measured by averaging the LL% values over the last minute of session two training, just before the start of the cooldown period in which after-effect is measured. This data is shown for all participants in the successful correlation category in Table 2. In 85.7% of the successful correlation → controlled intervention cases, the error between the real-time effect and the target calculated by Equation 6 was <40%, reaching 2.4% at best. This 40% value is rather large, and even still the results of these sessions were often very precise. Therefore, more research must be done regarding the fine tuning of the proportional and integral gains of the controller, as well as the step length averaging time frames (20 seconds in this study). Equation 6 shows Error calculation for comparing real-time effect to target asymmetry.

Error ⁡ ( % ) = Live - Target Target * 1 ⁢ 0 ⁢ 0 ⁢ % [ Equation ⁢ ⁢ 6 ]

TABLE 2 Subject 3 4 5 6 10 13 14 Baseline (%) 6.8 6.1 −0.1 −10.3 −4.0 8.8 12.6 Target (%) 15.4 17.9 −0.2 −23.1 −7.0 25.9 289.2 LiveEffect (%) 11.9 18.3 0.7 −14.1 −4.7 17.7 45.1 Error (%) 23.1 2.3 525.0 38.9 32.2 31.7 84.4 AfterEffect (%) 0.4 3.1 −0.4 −0.4 0.9 2.9 0.7 Subject 2 7 8 12 Baseline (%) 21.6 −15.9 25.8 9.4 Target (%) 1462.2 −20.7 259.6 127.7 LiveEffect (%) 34.0 −24.8 18.6 9.9 Error (%) 97.7 20.2 92.8 92.2 AfterEffect (%) 7.8 −6.9 18.0 5.0

Optimizing these values would likely result in tighter control of gait and allow for even more precise intervention. However, humans are often unpredictable in nature, and therefore achieving target asymmetries perfectly is unlikely. Ultimately, the most important aspect is continually fighting against the participant's natural tendency to adapt and return to their baseline gait.

3.6. Conclusions

Whenever the control software was able to operate within a reasonable range, it succeeded in a symmetric after-effect with a success rate of 85.7%. However, this number does not represent the whole picture, as many participants did not allow the controller to stay within a reasonable range. Exactly 50% of those who participated in this study exhibited an after-effect of <4% after only two sessions. The stark contrast between the success rate of the controlled intervention and the success rate of the entire study shows that, while the controller is highly accurate, more intricate therapy plans are necessary to use the feedback effectively. Ultimately, the newfound ability to accurately control and predict gait parameters expands the horizon of rehabilitation engineering. This strategy may be applied to a wide variety of gait parameters to produce individualized therapies for optimized rehabilitation. Further, the ability to determine participant's training session outcomes (session two) from their baseline session outcomes (session one) with reasonable accuracy provides great clinical benefit. In theory, several baseline sessions of varying real-time controlled therapies, such as split-belt and rhythmic cueing, could be performed to determine patient response to each of these interventions individually. These baseline sessions would allow for determination of which therapies the patient is sensitive to and which they are non-responsive to. When the controller was able to function, it did so with great success rates. Thus, the formation of a large dataset for a wide variety of therapies would give researchers the ability to quickly and efficiently pinpoint the necessary and highly effective interventions required by an individual.

FIG. 24 depicts a schematic block diagram of a computing device 2400 that can be used to implement various embodiments of the present disclosure. An exemplary computing device 2400 includes at least one processor circuit, for example, having a processor 2402 and a memory 2404, both of which are coupled to a local interface 2406, and one or more input and output (I/O) devices 2408. The local interface 2406 may comprise, for example, a data bus with an accompanying address/control bus or other bus structure as can be appreciated. The computing device 2400 further includes Graphical Processing Unit(s) (GPU) 2410 that are coupled to the local interface 2406 and may utilize memory 2404 and/or may have its own dedicated memory. The CPU and/or GPU(s) can perform various operations such as image enhancement, graphics rendering, image/video processing, recognition (e.g., text recognition, object recognition, feature recognition, etc.), image stabilization, machine learning, filtering, image classification, and any of the various operations described herein.

Stored in the memory 2404 are both data and several components that are executable by the processor 2402. In particular, stored in the memory 2404 and executable by the processor 2402 are code for measuring gait parameters 2407 as training feedback, code for gait perception feedback analysis logic/instructions 2409, code for adjusting an amount of therapy applied in a training session 2412 based on training feedback, and/or code for implementing one or more convolutional neural network (CNN) models 2411. Further, stored in the memory 2404 and executable by the processor 2402 are code for measuring one or more gait parameters of an individual, code for calculating a joint metric based on the measured one or more gait parameters, code for signaling an adjustment of an amount of one or more perturbation techniques based on the joint metric for reducing asymmetry in a gait pattern of the individual, iterating the measuring the one or more gait parameters of an individual, changing weight one foot adding weight to a leg of the individual, code for changing height on foot adding height to a leg of the individual, code for laterally pulling at hip laterally pulling at a hip of the individual, code for anteriorly pulling on foot anteriorly pulling on a leg of the individual, code for training with a split-belt treadmill having different speeds for two treads, code for measuring a left step length and a right step length, code for measuring an average step length for a predetermined period of time, calculating a left leg long percentage based on the left step length, the right step length, and the average step length, code for comparing the left leg long percentage with a predetermined left leg long percentage, and/or code for signaling an adjustment of amounts of speeds for two corresponding treads of a treadmill. In some examples, the computing device 2400 may be included in a split-belt treadmill, a mobile device.

Also stored in the memory 2404 may be a data store 2414 and other data. The data store 2414 can include an image database for gait videos, parameters, subjective ratings, and potentially other data. In addition, an operating system may be stored in the memory 2404 and executable by the processor 2402. The I/O devices 2408 may include input devices, for example but not limited to, a keyboard, touchscreen, mouse, one or more cameras and/or sensors, etc. For example, the sensors may include optical means (camera or depth/distance sensor), a wearable motion sensor, a wearable device (a simple phone in a pocket would probably be enough), a vibration/strike sensor integrated in a shoe or into a treadmill, a force plate, a motion capture camera, or any other suitable sensor. Furthermore, the I/O devices 2408 may also include output devices, for example but not limited to, speaker, earbuds, audio output port, a printer, display, Bluetooth output module, etc. In further examples, a treadmill may include controllable tension on lateral/foot pulls, which are controllable by the computing device 2400.

Certain embodiments of the present disclosure can be implemented in hardware, software, firmware, or a combination thereof. If implemented in software, the multi-initialization ensemble-based analysis logic or functionality are implemented in software or firmware that is stored in a memory and that is executed by a suitable instruction execution system. If implemented in hardware, the multi-initialization ensemble-based analysis logic or functionality can be implemented with any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit(s) having logic gates for implementing logic functions upon data signals, an application specific integrated circuit (ASIC) having appropriate combinational logic gates, a programmable gate array(s) (PGA), a field programmable gate array (FPGA), etc.

FIG. 25 is a flow chart illustrating an exemplary process 2500 for reducing gait asymmetry in accordance with some aspects of the present disclosure. As described below, a particular implementation may omit some or all illustrated features, and may not require some illustrated features to implement all embodiments. In some examples, the apparatus or computer device 2400 illustrated in FIG. 24 may be configured to carry out the process 2500. In some examples, any suitable apparatus or means for carrying out the functions or algorithm described below may carry out the process 2500.

At block 2502, the method can measure one or more gait parameters. In some scenarios, one or more gait parameters can be measured for a predetermined period of time using at least one of a force plate, a motion sensor, or a motion capture camera. For example, the one or more gait parameters are measured as an average of the one or more gait parameters for a predetermined period of time. For example, an average of one or more gait parameters can be measured for the last 20 seconds for a 5-minute session with one or more perturbation techniques. For example, the one or more gait parameters include at least one of: a spatiotemporal gait parameter, a kinetic gait parameter, or a force gait parameter. In some instances, the spatiotemporal gait parameter can include at least one of: a step length of the individual, a step time of the individual, a double limb support time of the individual, a stance time of the individual, or a swing time of the individual. In some examples, the step length may indicate the distance between the left and right heel when the front leg touches the ground. The step time may indicate the time between one foot's heel touching the ground the other foot touching the ground. The double limb support time may indicate the time that both legs are touching the ground. The above three parameters are defined in terms of the leading leg. For example, right step length is the length between heels when the right leg is in front. In further examples, the stance time may indicate the time that one foot is in contact with the ground. The swing time may indicate the time that one leg is in the air. The above two parameters are defined as the leg that is in stance or the leg that is in swing. In addition to the above parameters, the asymmetry of each of those parameters can be used. For example, step length asymmetry is the difference between left step length and right step length.

In further instances, the kinetic gait parameter can include at least one of: a hip angle of the individual, a knee angle of the individual, or an ankle angle of the individual. In some examples, the kinetic gait parameter may be measured using, for example, an accelerometer, optical tracking device, etc. In some examples, the hip angle may indicate the angle at the hip, which is measured as the difference between the torso and the thigh. This has a left and right hip angle that corresponds to the left and right thigh, respectively. The knee angle may indicate the angle at the knee, which is measured as the difference between the thigh and the shank of the leg. This has a left and right corresponding to the respective legs. The ankle angle may indicate the angle at the ankle, which is measured as the difference between the shank and the foot. This has a left and right corresponding to the respective legs. In addition to the above parameters, the asymmetry of each of those parameters can be used.

In further instances, the force gait parameter can include at least one of: a vertical force of the individual, a braking force of the individual, or a push-off force of the individual. In some examples, the force gait parameter may be measured using, for example, a force plate. In some examples, the vertical force may indicate the force that an individual exerts on ground in the direction of gravity (i.e., down). This has a left and right corresponding to the respective legs. The braking force may indicate the horizontal force that the person exerts on the ground that occurs when their foot first comes in contact with the ground at the start of stance phase. This has a left and right corresponding to the respective legs. The push-off force may indicate the force exerted by the foot at the end of stance phase right before initiating swing phase. This has a left and right corresponding to the respective legs. In addition to the above parameters, the asymmetry of each of those parameters can be used. In further examples, the one or more gait parameters may further include a gait perception parameter for the gait pattern of the individual. In the examples, the gait perception parameter may be considered as a weight parameter.

At block 2504, the method can calculate a joint metric based on the measured one or more gait parameters. A joint metric may indicate an overall level of asymmetry of the gait pattern. E.g., CGAM. For example, the joint metric is calculated by incorporating the measured one or more gait parameters for indicating an incorporated level of asymmetry in the gait pattern of the individual. In some examples, the joint metric is a single representation of the measured gait parameters that generally scales with the global deviation from symmetry and indicates how far away a gait is from an ideal. The deviation of each measure is scaled based on the variance within that measure, so measures that generally have larger magnitudes of asymmetry (e.g., forces) will be scaled so that each gait parameter has a similar influence on the overall metric. These weightings are a starting point and the weight assigned to each metric can be evaluated and can be adjusted based on the effect of each parameter and based on the biomechanics related to the gait parameters.

At block 2506, the method can signal an adjustment of an amount of one or more perturbation techniques based on the joint metric for reducing asymmetry in a gait pattern of the individual. In some examples, the adjustment is signaled based on real-time feedback to measure the one or more gait parameters. In some examples, the one or more perturbation techniques applies physical properties of the individual or external manipulations that affect the one or more gait parameters. For example, the one or more perturbation techniques may include asymmetric rhythmic auditory cueing making a first audible cue with a first cue duration for one leg of the individual and a second audible cue with a second cue duration for another leg of the individual. Here, the first cue duration may be different from the second cue duration. The one or more perturbation techniques may further include at least one of: changing weight one foot adding weight to a leg of the individual, changing height on foot adding height to a leg of the individual, lateral pulling at hip laterally pulling at a hip of the individual, anterior pulling on foot anteriorly pulling on a leg of the individual, or split-belt treadmill training having different speeds for two treads. In further examples, the one or more perturbation techniques may include performing split-belt treadmill having different speeds for two treads. In the examples, a first adjustment of the asymmetric rhythmic auditory cueing and a second adjustment of split-belt treadmill training can be simultaneously signaled. In the examples, the first adjustment of the asymmetric rhythmic auditory cueing may be an auditory cueing ratio of 2:1. The second adjustment of split-belt treadmill training may be a split-belt ratio of 2:1. In some examples, The auditory cueing ratio is the relation between two cues where the number of times one value contains or is contained within the other. The ratio defines the relative delay between the right cue playing and the left cue. This ratio can be any amount, but is typically between 5:1 and 1:5. This change will focus on step time. The stride time defines the time taken for the left and right cues to both be played. The split belt ratio the relation between the speeds of the two belts in which the left and right legs are waling. The ratio defines the relative speed difference between the right belt speed and the left belt speed. This change will focus on step length. This ratio can be any amount, but is typically between 5:1 and 1:5

At block 2508, the method can determine whether to repeat the processes at blocks 2502-2506. For example, the method can iterate the measuring the one or more gait parameters of an individual, the calculating the joint metric based on the measured one or more gait parameters, and the signaling the adjustment of the amount of one or more perturbation techniques based on the joint metric.

FIG. 26 is a flow chart illustrating an exemplary process 2600 for reducing gait asymmetry in accordance with some aspects of the present disclosure. As described below, a particular implementation may omit some or all illustrated features, and may not require some illustrated features to implement all embodiments. In some examples, the apparatus or computer device 2400 illustrated in FIG. 24 may be configured to carry out the process 2600. In some examples, any suitable apparatus or means for carrying out the functions or algorithm described below may carry out the process 2600.

At block 2602, the method can measure a left gait parameter and a right gait parameter. In some examples, the gait parameter may be a step time and/or a step length. Thus, the left gait parameter may include a left step time and/or a left step length. The right gait parameter may include a right step time and/or a right step length

At block 2604, the method can measure an average of the left and right step gait parameters for a predetermined period of time. In some examples, the average may include an average of the first step time and the second step time and/or an average of the first step length and the second step length.

At block 2606, the method can calculate a left leg long percentage based on the left gait parameter, the right gait parameter and the average of the left and right parameters. In some examples, the left leg long percentage may be calculated based on the left step length, the right step length, and the average step length of the left and right step lengths. For example, the left leg long percentage is calculated by: the left leg long percentage=(the left step length−the right step length)/(the average step length)×100. In other examples, the left leg long percentage may be calculated based on the left step time, the right step time, and the average step time of the left and right step times. For example, the left leg long percentage is calculated by: the left leg long percentage=(the left step time−the right step time)/(the average step time)×100.

At block 2608, the method can compare the left leg long percentage with a predetermined left leg long percentage.

At block 2610, the method can signal an adjustment of amounts of asymmetry based on the compared left leg long percentage. In some examples, the method can signal the adjustment of tread speeds based on the compared left leg long percentage for the for two corresponding treads of a treadmill. In some examples, the adjustment is signaled with real-time feedback to measure the left step length/time and right step length/time and measure the average step length/time. In further examples, a ratio of the amounts of asymmetry for two corresponding treads is equal to or less than a split-belt ratio of 5:1 or 1:5. For example, the ratio of the tread speeds can be any value between 1:5 and 5:1 as adjusted by the feedback controller. In other examples, the method can signal the adjustment of cue duration based on the compared left leg long percentage for two corresponding times between cues The ratio of the time between cues can be any value between 1:5 and 5:1 as adjusted by the feedback controller.

At block 2604, the method can determine whether to repeat the processes 2602-2610.

It should be emphasized that the above-described embodiments of the present disclosure are merely possible examples of implementations, merely set forth for a clear understanding of the principles of the disclosure. Many variations and modifications may be made to the above-described embodiment(s) without departing substantially from the principles of the disclosure. All such modifications and variations are intended to be included herein within the scope of this disclosure and protected by the following claims.

Claims

1. A method for reducing gait asymmetry comprising:

measuring one or more gait parameters of an individual;
calculating a joint metric based on the measured one or more gait parameters; and
signaling an adjustment of an amount of one or more perturbation techniques based on the joint metric for reducing asymmetry in a gait pattern of the individual, wherein one or more perturbation techniques comprise asymmetric rhythmic auditory cueing making a first audible cue with a first cue duration for one leg of the individual and a second audible cue with a second cue duration for another leg of the individual, the first cue duration being different from the second cue duration.

2. The method of claim 1, wherein the signaling the adjustment is real-time feedback to measure the one or more gait parameters.

3. The method of claim 1, wherein the one or more gait parameters comprise at least one of: a spatiotemporal gait parameter, a kinetic gait parameter, or a force gait parameter.

4. The method of claim 1, wherein the spatiotemporal gait parameter comprises at least one of: a step length of the individual, a step time of the individual, a double limb support time of the individual, a stance time of the individual, or a swing time of the individual.

5. The method of claim 1, wherein the kinetic gait parameter comprises at least one of: a hip angle of the individual, a knee angle of the individual, or an ankle angle of the individual.

6. The method of claim 1, wherein the force gait parameter comprises at least one of: a vertical force of the individual, a braking force of the individual, or a push-off force of the individual.

7. The method of claim 1, wherein the measuring one or more gait parameters comprises measuring the one or more gait parameters for a predetermined period of time using at least one of a force plate or a motion capture camera.

8. The method of claim 1, wherein the joint metric is calculated by incorporating the measured one or more gait parameters for indicating an incorporated level of asymmetry in the gait pattern of the individual.

9. The method of claim 1, further comprising:

iterating the measuring the one or more gait parameters of an individual, the calculating the joint metric based on the measured one or more gait parameters, and the signaling the adjustment of the amount of one or more perturbation techniques based on the joint metric.

10. The method of claim 1, wherein the one or more perturbation techniques applies physical properties of the individual or external manipulations that affect the one or more gait parameters.

11. The method of claim 1, wherein the one or more perturbation techniques further comprises at least one of:

changing weight one foot adding weight to a leg of the individual;
changing height on foot adding height to a leg of the individual;
laterally pulling at hip laterally pulling at a hip of the individual;
anteriorly pulling on foot anteriorly pulling on a leg of the individual; or
training with a split-belt treadmill having different speeds for two treads.

12. The method of claim 1, wherein the one or more gait parameters further comprise a gait perception parameter for the gait pattern of the individual.

13. The method of claim 12, wherein the gait perception parameter is a weight parameter.

14. The method of claim 1, wherein the one or more perturbation techniques further comprises performing split-belt treadmill having different speeds for two treads, and

wherein the signaling the adjustment comprises simultaneously signaling a first adjustment of the asymmetric rhythmic auditory cueing and a second adjustment of split-belt treadmill training.

15. The method of claim 14, wherein the first adjustment of the asymmetric rhythmic auditory cueing is an auditory cueing ratio of 2:1, and

wherein the second adjustment of split-belt treadmill training is a split-belt ratio of 2:1.

16. A method for reducing gait asymmetry comprising:

measuring a left step gait parameter and a right step gait parameter;
measuring an average of the left and right step gait parameters for a predetermined period of time;
calculating a left leg long percentage based on the left gait parameter, the right gait parameter, and the average;
comparing the left leg long percentage with a predetermined left leg long percentage; and
signaling an adjustment of amounts of asymmetry based on the compared left leg long percentage.

17. The method of claim 16, wherein the signaling the adjustment of amounts of asymmetry comprising signaling the adjustment of amounts of tread speeds based on the compared left leg long percentage for two corresponding treads of a treadmill.

18. The method of claim 17, wherein the signaling the adjustment is real-time feedback to measure the left gait parameter and right gait parameter and measure the average.

19. The method of claim 17, wherein the left leg long percentage is calculated by: the ⁢ ⁢ left ⁢ ⁢ leg ⁢ ⁢ long ⁢ ⁢ percentage ⁢ = the ⁢ ⁢ left ⁢ ⁢ gait ⁢ ⁢ parameter - the ⁢ ⁢ right ⁢ ⁢ gait ⁢ ⁢ parameter the ⁢ ⁢ average × 1 ⁢ 0 ⁢ 0.

20. The method of claim 17, wherein a ratio of the amounts of asymmetry for two corresponding treads is between a split-belt ratio of 5:1 and 1:5.

21. The method of claim 16, wherein the signaling the adjustment of amounts of asymmetry comprising signaling the adjustment of cue duration based on the compared left leg long percentage for two corresponding times between cues.

22. The method of claim 21, wherein a ratio of the amounts of times for two corresponding cues is between a cue ratio of 5:1 and 1:5.

23. An apparatus for reducing gait asymmetry, comprising:

a processor; and
a memory communicatively coupled to the processor,
wherein the processor and the memory are configured to: measure one or more gait parameters of an individual; calculate a joint metric based on the measured one or more gait parameters; and signal an adjustment of an amount of one or more perturbation techniques based on the joint metric for reducing asymmetry in a gait pattern of the individual, wherein one or more perturbation techniques comprise asymmetric rhythmic auditory cueing making a first audible cue with a first cue duration for one leg of the individual and a second audible cue with a second cue duration for another leg of the individual, the first cue duration being different from the second cue duration.

24. The apparatus of claim 23, wherein the signaling the adjustment is real-time feedback to measure the one or more gait parameters.

25. The apparatus of claim 23, wherein the one or more gait parameters comprise at least one of: a spatiotemporal gait parameter, a kinetic gait parameter, or a force gait parameter.

26. The apparatus of claim 23, wherein the spatiotemporal gait parameter comprises at least one of: a step length of the individual, a step time of the individual, a double limb support time of the individual, a stance time of the individual, or a swing time of the individual.

27. The apparatus of claim 23, wherein the kinetic gait parameter comprises at least one of: a hip angle of the individual, a knee angle of the individual, or an ankle angle of the individual.

28. The apparatus of claim 23, wherein the force gait parameter comprises at least one of: a vertical force of the individual, a braking force of the individual, or a push-off force of the individual.

29. The apparatus of claim 23, wherein the measuring one or more gait parameters comprises measuring the one or more gait parameters for a predetermined period of time using at least one of a force plate or a motion capture camera.

30. The apparatus of claim 23, wherein the joint metric is calculated by incorporating the measured one or more gait parameters for indicating an incorporated level of asymmetry in the gait pattern of the individual.

31. The apparatus of claim 23, wherein the processor and the memory are further configured to:

iterate the measuring the one or more gait parameters of an individual, the calculating the joint metric based on the measured one or more gait parameters, and the signaling the adjustment of the amount of one or more perturbation techniques based on the joint metric.

32. The apparatus of claim 23, the one or more perturbation techniques applies physical properties of the individual or external manipulations that affect the one or more gait parameters.

33. The apparatus of claim 23, wherein the one or more perturbation techniques further comprises at least one of:

changing weight one foot adding weight to a leg of the individual;
changing height on foot adding height to a leg of the individual;
laterally pulling at hip laterally pulling at a hip of the individual;
anteriorly pulling on foot anteriorly pulling on a leg of the individual; or
training with a split-belt treadmill having different speeds for two treads.

34. The apparatus of claim 23, wherein the one or more gait parameters further comprise a gait perception parameter for the gait pattern of the individual.

35. The apparatus of claim 34, wherein the gait perception parameter is a weight parameter.

36. The apparatus of claim 23, wherein the one or more perturbation techniques further comprises performing split-belt treadmill having different speeds for two treads, and

wherein the signaling the adjustment comprises simultaneously signaling a first adjustment of the asymmetric rhythmic auditory cueing and a second adjustment of split-belt treadmill training.

37. The apparatus of claim 36, wherein the first adjustment of the asymmetric rhythmic auditory cueing is an auditory cueing ratio of 2:1, and

wherein the second adjustment of split-belt treadmill training is a split-belt ratio of 2:1.

38. An apparatus for reducing gait asymmetry, comprising:

a processor; and
a memory communicatively coupled to the processor,
wherein the processor and the memory are configured to: measure a left gait parameter and a right gait parameter; measure an average of the left and right step gait parameters for a predetermined period of time; calculate a left leg long percentage based on the left gait parameter, the right gait parameter, and the average; compare the left leg long percentage with a predetermined left leg long percentage; and signal an adjustment of amounts of asymmetry for two corresponding treads of a treadmill based on the compared left leg long percentage.

39. The apparatus of claim 38, wherein the signaling the adjustment of amounts of asymmetry comprising signaling the adjustment of amounts of tread speeds based on the compared left leg long percentage for two corresponding treads of a treadmill.

40. The apparatus of claim 39, wherein the signaling the adjustment is real-time feedback to measure the left gait parameter and right gait parameter and measure the average.

41. The apparatus of claim 39, wherein the left leg long percentage is calculated by: the ⁢ ⁢ left ⁢ ⁢ leg ⁢ ⁢ long ⁢ ⁢ percentage ⁢ = the ⁢ ⁢ left ⁢ ⁢ gait ⁢ ⁢ parameter - the ⁢ ⁢ right ⁢ ⁢ gait ⁢ ⁢ parameter the ⁢ ⁢ average × 1 ⁢ 0 ⁢ 0.

42. The apparatus of claim 39, wherein a ratio of the amounts of asymmetry for two corresponding treads is between a split-belt ratio of 5:1 and 1:5.

43. The apparatus of claim 38, wherein the signaling the adjustment of amounts of asymmetry comprising signaling the adjustment of cue duration based on the compared left leg long percentage for two corresponding times between cues.

44. The apparatus of claim 43, wherein a ratio of the amounts of times for two corresponding cues is between a cue ratio of 5:1 and 1:5.

Patent History
Publication number: 20220126191
Type: Application
Filed: Oct 26, 2021
Publication Date: Apr 28, 2022
Applicant: University of South Florida (Tampa, FL)
Inventors: KYLE REED (TAMPA, FL), SEOK HUN KIM (TAMPA, FL)
Application Number: 17/511,332
Classifications
International Classification: A63B 71/06 (20060101); A63B 22/02 (20060101); A63B 24/00 (20060101);