SYSTEMS AND METHODS FOR CLOSED-LOOP DETERMINATION OF STIMULATION PARAMETER SETTINGS FOR AN ELECTRICAL SIMULATION SYSTEM
A method or system for facilitating the determining and setting of stimulation parameters for programming an electrical stimulation system using closed loop programming is provided. For example, pulse generator feedback logic is executed by a processor to interface with control instructions of an implantable pulse generator by incorporating one or more machine learning engines to automatically generate a proposed set of stimulation parameter values that each affect a stimulation aspect of the implantable pulse generator, receive one or more clinical responses and automatically generate a revised set of values taking into account the received clinical responses, and repeating the automated receiving of a clinical response and adjusting the stimulation parameter values taking the clinical response into account, until or unless a stop condition is reach or the a therapeutic response is indicated within a designated tolerance.
This application is a continuation of U.S. patent application Ser. No. 16/748,161, filed Jan. 21, 2020, which is a continuation of U.S. patent application Ser. No. 15/783,961, filed Oct. 13, 2017, which issued as U.S. Pat. No. 10,603,498, which claims the benefit under 35 U.S.C. § 119(e) of U.S. Provisional Patent Application Ser. No. 62/408,620, filed Oct. 14, 2016, and U.S. Provisional Patent Application Ser. No. 62/545,819, filed Aug. 15, 2017, all of which are incorporated herein by reference.
FIELDThe present invention is directed to the area of implantable electrical stimulation systems and methods of making and using the systems. The present invention is also directed to systems and methods for facilitating determining and setting of stimulation parameters for programming an electrical stimulation system.
BACKGROUNDImplantable electrical stimulation systems have proven therapeutic in a variety of diseases and disorders. For example, spinal cord stimulation systems have been used as a therapeutic modality for the treatment of chronic pain syndromes. Peripheral nerve stimulation has been used to treat chronic pain syndrome and incontinence, with a number of other applications under investigation. Functional electrical stimulation systems have been applied to restore some functionality to paralyzed extremities in spinal cord injury patients. Stimulation of the brain, such as deep brain stimulation, can be used to treat a variety of diseases or disorders.
Stimulators have been developed to provide therapy for a variety of treatments. A stimulator can include a control module (with a pulse generator), one or more leads, and an array of stimulator electrodes on each lead. The stimulator electrodes are in contact with or near the nerves, muscles, or other tissue to be stimulated. The pulse generator in the control module generates electrical pulses that are delivered by the electrodes to body tissue.
BRIEF SUMMARYOne embodiment is a system facilitating programming settings of an implantable pulse generator associated with a patient using closed loop programming, the system including a computer processor and a memory which stores a pulse generator feedback control logic. The pulse generator feedback logic is executed by the processor to interface with control instructions of an implantable pulse generator by the following actions: incorporating one or more machine learning engines, automatically generating a proposed set of stimulation parameter values that each affect a stimulation aspect of the IPG; forwarding the automatically generated proposed set of stimulation parameter values to configure stimulation parameters of the IPG accordingly; receiving one or more clinical response values as a result of the IPG being configured to the proposed set of stimulation parameter values; automatically generating, by incorporating one or more machine learning engines, a revised proposed set of stimulation parameter values taking into account the received clinical response values; forwarding the revised proposed set to configure stimulation parameters of the IPG to the revised proposed set of stimulation parameter values; and repeating the receiving one or more clinical response values as a result of the IPG being configured to the revised proposed set, automatically generating, by incorporating one or more machine learning engines, a revised proposed set of stimulation parameter values, and forwarding the revised prosed set of stimulation parameters to configure stimulation parameters of the IPG accordingly, until or unless a stop condition has been reached or the one or more received clinical response values indicates a value that corresponds to a therapeutic response indication within a designated tolerance.
In at least some embodiments of the system, the IPG is a deep brain stimulator.
In at least some embodiments of the system, the stimulation parameter values that each affect a stimulation aspect of the IPG include one or more of: electrode determination, amplitude of current, frequency of current, pulse width, spread and/or foci of stimulation, and/or geometry of stimulated tissue
In at least some embodiments of the system, at least one stimulation parameter indicates a location of an electrode disposed on a lead of the electrical stimulator. In some of these embodiments, the electrode is a ring electrode or a segmented electrode.
In at least some embodiments of the system, at least one stimulation parameter indicates a location of an electrode that results in current being confined to a single direction.
In at least some embodiments of the system, the one or more clinical response values are automatically received from a sensor associated with the patient. In at least some of these embodiments, receiving the clinical response values from the sensors allows the programming of the IPG to be automated or at least semi-automated so that a clinician does not need to be present to program the IPG.
In at least some embodiments of the system, the pulse generator feedback control logic incorporates multiple machine learning techniques to automatically generate the proposed set of stimulation parameter values and/or the revised proposed set of stimulation parameter values. In at least some of these embodiments the machine learning techniques are deployed simultaneously. In at least some embodiments the multiple learning techniques are rated (or ranked) automatically based upon ability to generate a set of stimulation parameter values that corresponds to the therapeutic response indication within the designated tolerance or based upon other metrics. These ratings may change within a stimulation parameter setting or between different sessions.
In at least some embodiments of the system, the automatically generating, by incorporating one or more machine learning engines, a revised proposed set of stimulation parameter values further comprises, using the one or more machine learning engines, automatically generating one or more predicted therapeutic response values for a plurality of values of the plurality of stimulation parameters of the IPG for which a clinical response value has not yet been received; and automatically selecting a revised proposed set of stimulation parameter values to configure the IPG based in part on at least one of the predicted therapeutic response values. In at least some of these embodiments, multiple machine learning engines are deployed to generate multiple predications of the one or more predicted therapeutic response values. In at least some of these embodiments, the revised proposed set of stimulation parameter values to configure the IPG takes into account at least one of the predicted therapeutic response values to determine a next set of stimulation parameter values to try.
In at least some embodiments of the system, the automatically generating the revised proposed set of values of the plurality of stimulation parameters is based upon one or more rules. In at least some of these embodiments, the rules rank the remaining possible proposed sets of values. In at least some of these embodiments, the rules include at least one of: a rule based upon retrying locations within a number of iterations, a rule based upon mathematical proximity of proposed sets of stimulation parameter values to sets of stimulation parameter values already tested, a rule based upon values of a stimulation parameter leading to adverse side effects, and/or a rule based upon size of step between values of a same stimulation parameter.
In at least some embodiments of the system, the receiving one or more clinical response values as a result of the IPG being configured to the proposed set of stimulation parameter values indicates whether an undesired side effect has occurred.
In at least some embodiments of the system, the revised proposed set of stimulation parameter values are determined in each subsequent iteration in contact order, wherein contact order means testing along a first contact until side effects are prohibitive before testing a next proximal contact on a lead. In at least some embodiments of the system, the revised proposed set of stimulation parameter values are determined in each subsequent iteration by selectively choosing contacts along the main axis of a lead and varying other stimulation parameters for each selectively chosen contact.
In at least some embodiments of the system, the revised proposed set of stimulation parameter values are determined in each subsequent iteration randomly, based upon a model of the patient, a spatial search algorithm, or by varying size of jump to a revised proposed set of stimulation parameter values. In other embodiments, the revised proposed set of stimulation parameter values may be determined using other techniques or rules.
Another embodiment is a method for facilitating programming settings of an electrical stimulator associated with a patient using closed loop programming that includes: receiving an indication of a plurality of stimulation parameters and an indication of a desired therapeutic response value; automatically generating an initial proposed set of stimulation parameter values for testing the electrical stimulator programmed according to these initial values; receiving an indication of clinical results based upon testing the programmed electrical stimulator on a recipient, wherein the received indication indicates a therapeutic response value; automatically generating, using machine learning, one or more predicted therapeutic response values for a plurality of values of the plurality of stimulation parameters for which an indication of clinical results has not yet been received; automatically generating and indicating a next proposed set of stimulation parameter values for testing the electrical stimulator programmed accordingly, based in part on at least one of the predicted therapeutic response values; and repeating the acts of receiving the indication of clinical results based upon testing the programmed electrical simulator, automatically generating using machine learning the one or more predicted therapeutic response values, and automatically generating and indicating a next proposed set of stimulation parameter values for testing the electrical stimulator programmed accordingly, until or unless a stop condition has been reached or the received indication of clinical results indicates a therapeutic response value that is within a designated tolerance.
In some embodiments of the method, electrical stimulator is an implantable pulse generator. In some embodiments of the method, at least one stimulation parameter indicates a location of an electrode disposed on a lead of the electrical stimulator. In some embodiments of the method, the electrode is a ring electrode or a segmented electrode. In some embodiments of the method, at least one stimulation parameter results in current stimulation in multiples directions at once. In some embodiments of the method, at least one stimulation parameter indicates a location of an electrode that results in current being confined to a single direction.
In at least some embodiments of the method, the plurality of stimulation parameter values includes one or more of: electrode location, amplitude of current, frequency of current, pulse width, spread of stimulation, and/or geometry of stimulated tissue.
In at least some embodiments of the method, the receiving the indication of clinical results based upon testing the recipient with an electrical stimulator indicates whether an undesired side effect has occurred.
In at least some embodiments of the method, the automatically generating, using machine learning, one or more predicted therapeutic response values for a plurality of values of the plurality of stimulation parameters for which an indication of clinical results has not yet been received uses a training data set based upon prior data of the recipient or of other recipients.
In at least some embodiments of the method, the indication of clinical results based upon testing the electrical stimulator on a recipient is received from a data repository.
In at least some embodiments of the method, the initial proposed set of stimulation parameter values is determined using data collected in prior electrical simulator programming sessions.
In at least some embodiments of the method, multiple machine learning techniques are deployed simultaneously to generate multiple predictions of the one or more predicted therapeutic response values and the multiple machine learning techniques are rated automatically based upon ability to provide an accurate prediction of a therapeutic response measurement.
In at least some embodiments of the method, the automatically generating a next proposed set of stimulation parameter values determines the next proposed set by ranking possible sets of stimulation parameter values based upon one or more rules. In at least some of these embodiments, the rules rank remaining possible proposed sets of values. In at least some of these embodiments, the rules include at least one of: a rule based upon retrying locations within a number of iterations, rule based upon mathematical proximity of proposed sets of stimulation parameter values to sets of stimulation parameter values already tested, a rule based upon values of a stimulation parameter leading to adverse side effects, and/or a rule based upon size of step between values of a same stimulation parameter.
In at least some embodiments of the method, the next proposed set of stimulation parameter values is determined in each subsequent iteration in contact order, wherein contact order means testing along a first contact until side effects are prohibitive before testing a next proximal contact on a lead. In at least some embodiments of the method, the next proposed set of stimulation parameter values is determined in each subsequent iteration by selectively choosing contacts along the main access of a lead and varying other stimulation parameters for each selectively chosen contact. In at least some embodiments of the method, the next proposed set of stimulation parameter values is determined in each subsequent iteration randomly, based upon a model of the patient, or a spatial search algorithm. In at least some embodiments of the method, the next proposed set of stimulation parameter values is determined in each subsequent iteration by varying size of jump to a next proposed set of combined values of the plurality of stimulation parameters.
In at least some embodiments of the method, value limits are defined for at least some of the plurality of stimulation parameters.
In at least some embodiments of the method, the method is considered to have determined an optimal response based upon fastest time to determine: a therapeutic response value that is within the designated tolerance, a therapeutic response value that is closest to desired therapeutic response value, or least battery use for some therapeutic response value that is within the designated tolerance.
Yet another embodiment is a computer-readable medium having instructions stored thereon that includes any of the acts of the pulse generator feedback control logic of any of the system embodiments described or that includes any of the acts of any of the method embodiments described.
In at least one embodiment, the computer-readable medium includes instructions that control a computer processor to automatically determine programming settings of an implantable pulse generator (IPG) associated with a patient by performing a method including: automatically generating a proposed set of values of a plurality of stimulation parameters that each affect a stimulation aspect of the IPG; causing stimulation parameters of the IPG to be set to the automatically generated proposed set of stimulation parameter values; receiving one or more clinical response values as a result of the IPG being set to the proposed set of stimulation parameter values; automatically generating, by incorporating one or more machine learning engines, a revised proposed set of stimulation parameter values; causing stimulation parameters of the IPG to be set to the automatically generated revised proposed set of stimulation parameter values; and repeating the receiving the one or more clinical response values, automatically generating, by incorporating one or more machine learning engines, a revised proposed set of stimulation parameter values, and causing the stimulation parameters of the IPG to be set to the revised proposed set of stimulation parameter values, until or unless a stop condition has been reached or the received one or more clinical response values indicates a value that corresponds to a therapeutic response measurement within a designated tolerance.
In at least one embodiment, the medium is a memory in a computing system.
Non-limiting and non-exhaustive embodiments of the present techniques are described with reference to the following drawings. In the drawings, like reference numerals refer to like parts throughout the various figures unless otherwise specified.
For a better understanding of the present techniques, reference will be made to the following Detailed Description, which is to be read in association with the accompanying drawings, wherein:
The present invention is directed to the area of implantable electrical stimulation systems and methods of making and using the systems. The present invention is also directed to systems and methods for automatically determining advantageous stimulation parameter values using machine learning techniques to explore and/or set parameters of an electrical stimulation device.
Suitable implantable electrical stimulation systems include, but are not limited to, a least one lead with one or more electrodes disposed on a distal end of the lead and one or more terminals disposed on one or more proximal ends of the lead. Leads include, for example, percutaneous leads, paddle leads, cuff leads, or any other arrangement of electrodes on a lead. Examples of electrical stimulation systems with leads are found in, for example, U.S. Pat. Nos. 6,181,969; 6,516,227; 6,609,029; 6,609,032; 6,741,892; 7,244,150; 7,450,997; 7,672,734; 7,761,165; 7,783,359; 7,792,590; 7,809,446; 7,949,395; 7,974,706; 8,175,710; 8,224,450; 8,271,094; 8,295,944; 8,364,278; 8,391,985; and 8,688,235; and U.S. Patent Applications Publication Nos. 2007/0150036; 2009/0187222; 2009/0276021; 2010/0076535; 2010/0268298; 2011/0005069; 2011/0004267; 2011/0078900; 2011/0130817; 2011/0130818; 2011/0238129; 2011/0313500; 2012/0016378; 2012/0046710; 2012/0071949; 2012/0165911; 2012/0197375; 2012/0203316; 2012/0203320; 2012/0203321; 2012/0316615; 2013/0105071; and 2013/0197602, all of which are incorporated by reference. In the discussion below, a percutaneous lead will be exemplified, but it will be understood that the methods and systems described herein are also applicable to paddle leads and other leads.
A lead for electrical stimulation (for example, deep brain or spinal cord stimulation) includes stimulation electrodes that can be ring electrodes, segmented electrodes that extend only partially around the circumference of the lead, or any other type of electrode, or any combination thereof. The segmented electrodes can be provided in sets of electrodes, with each set having electrodes circumferentially distributed about the lead at a particular longitudinal position. For illustrative purposes, the leads, systems, and methods are described herein relative to use for deep brain stimulation, but it will be understood that any of the leads, systems, and methods can be used for applications other than deep brain stimulation, including spinal cord stimulation, peripheral nerve stimulation, dorsal root ganglion stimulation, or stimulation of other nerves, muscles, and tissues.
Turning to
The IPG 14 is physically connected, optionally via one or more lead extensions 24, to the stimulation lead(s) 12. Each lead carries multiple electrodes 26 arranged in an array. The IPG 14 includes pulse generation circuitry that delivers electrical stimulation energy in the form of, for example, a pulsed electrical waveform (i.e., a temporal series of electrical pulses) to the electrode array 26 in accordance with a set of stimulation parameters. The implantable pulse generator can be implanted into a patient's body, for example, below the patient's clavicle area or within the patient's buttocks or abdominal cavity. The implantable pulse generator can have eight or sixteen stimulation channels which may be independently programmable to control the magnitude of the current stimulus from each channel. In some embodiments, the implantable pulse generator can have more or fewer than eight or sixteen stimulation channels (e.g., 4-, 6-, 32-, or more stimulation channels). The implantable pulse generator can have one, two, three, four, or more connector ports, for receiving the terminals of the leads.
The ETS 20 may also be physically connected, optionally via the percutaneous lead extensions 28 and external cable 30, to the stimulation leads 12. The ETS 20, which may have similar pulse generation circuitry as the IPG 14, also delivers electrical stimulation energy in the form of, for example, a pulsed electrical waveform to the electrode array 26 in accordance with a set of stimulation parameters. One difference between the ETS 20 and the IPG 14 is that the ETS 20 is often a non-implantable device that is used on a trial basis after the neurostimulation leads 12 have been implanted and prior to implantation of the IPG 14, to test the responsiveness of the stimulation that is to be provided. Any functions described herein with respect to the IPG 14 can likewise be performed with respect to the ETS 20.
The RC 16 may be used to telemetrically communicate with or control the IPG 14 or ETS 20 via a uni- or bi-directional wireless communications link 32. Once the IPG 14 and neurostimulation leads 12 are implanted, the RC 16 may be used to telemetrically communicate with or control the IPG 14 via a uni- or bi-directional communications link 34. Such communication or control allows the IPG 14 to be turned on or off and to be programmed with different stimulation parameter sets. The IPG 14 may also be operated to modify the programmed stimulation parameters to actively control the characteristics of the electrical stimulation energy output by the IPG 14. The CP 18 allows a user, such as a clinician, the ability to program stimulation parameters for the IPG 14 and ETS 20 in the operating room and in follow-up sessions.
The CP 18 may perform this function by indirectly communicating with the IPG 14 or ETS 20, through the RC 16, via a wireless communications link 36. Alternatively, the CP 18 may directly communicate with the IPG 14 or ETS 20 via a wireless communications link (not shown). The stimulation parameters provided by the CP 18 are also used to program the RC 16, so that the stimulation parameters can be subsequently modified by operation of the RC 16 in a stand-alone mode (i.e., without the assistance of the CP 18).
For purposes of brevity, the details of the RC 16, CP 18, ETS 20, and external charger 22 will not be further described herein. Details of exemplary embodiments of these devices are disclosed in U.S. Pat. No. 6,895,280, which is expressly incorporated herein by reference. Other examples of electrical stimulation systems can be found at U.S. Pat. Nos. 6,181,969; 6,516,227; 6,609,029; 6,609,032; 6,741,892; 7,949,395; 7,244,150; 7,672,734; and 7,761,165; 7,974,706; 8,175,710; 8,224,450; and 8,364,278; and U.S. Patent Application Publication No. 2007/0150036, as well as the other references cited above, all of which are incorporated by reference.
In some embodiments, measurement devices coupled to the muscles or other tissues stimulated by the target neurons, or a unit responsive to the patient or clinician, can be coupled to the implantable pulse generator or microdrive motor system. The measurement device, user, or clinician can indicate a response by the target muscles or other tissues to the stimulation or recording electrode(s) to further identify the target neurons and facilitate positioning of the stimulation electrode(s). For example, if the target neurons are directed to a muscle experiencing tremors, a measurement device can be used to observe the muscle and indicate changes in, for example, tremor frequency or amplitude in response to stimulation of neurons. Alternatively, the patient or clinician can observe the muscle and provide feedback.
The lead 110 for deep brain stimulation can include stimulation electrodes, recording electrodes, or both. In at least some embodiments, the lead 110 is rotatable so that the stimulation electrodes can be aligned with the target neurons after the neurons have been located using the recording electrodes.
Stimulation electrodes may be disposed on the circumference of the lead 110 to stimulate the target neurons. Stimulation electrodes may be ring-shaped so that current projects from each electrode equally in every direction from the position of the electrode along a length of the lead 110. In the embodiment of
The lead 100 includes a lead body 110, terminals 145, and one or more ring electrodes 120 and one or more sets of segmented electrodes 130 (or any other combination of electrodes). The lead body 110 can be formed of a biocompatible, non-conducting material such as, for example, a polymeric material. Suitable polymeric materials include, but are not limited to, silicone, polyurethane, polyurea, polyurethane-urea, polyethylene, or the like. Once implanted in the body, the lead 100 may be in contact with body tissue for extended periods of time. In at least some embodiments, the lead 100 has a cross-sectional diameter of no more than 1.5 mm and may be in the range of 0.5 to 1.5 mm. In at least some embodiments, the lead 100 has a length of at least 10 cm and the length of the lead 100 may be in the range of 10 to 70 cm.
The electrodes 125 can be made using a metal, alloy, conductive oxide, or any other suitable conductive biocompatible material. Examples of suitable materials include, but are not limited to, platinum, platinum iridium alloy, iridium, titanium, tungsten, palladium, palladium rhodium, or the like. Preferably, the electrodes are made of a material that is biocompatible and does not substantially corrode under expected operating conditions in the operating environment for the expected duration of use.
Each of the electrodes can either be used or unused (OFF). When the electrode is used, the electrode can be used as an anode or cathode and carry anodic or cathodic current. In some instances, an electrode might be an anode for a period of time and a cathode for a period of time.
Deep brain stimulation leads and other leads may include one or more sets of segmented electrodes. Segmented electrodes may provide for superior current steering than ring electrodes because target structures in deep brain stimulation or other stimulation are not typically symmetric about the axis of the distal electrode array. Instead, a target may be located on one side of a plane running through the axis of the lead. Through the use of a radially segmented electrode array (“RSEA”), current steering can be performed not only along a length of the lead but also around a circumference of the lead. This provides precise three-dimensional targeting and delivery of the current stimulus to neural target tissue, while potentially avoiding stimulation of other tissue. Examples of leads with segmented electrodes include U.S. Pat. Nos. 8,473,061; 8,571,665; and 8,792,993; U.S. Patent Application Publications Nos. 2010/0268298; 2011/0005069; 2011/0130803; 2011/0130816; 2011/0130817; 2011/0130818; 2011/0078900; 2011/0238129; 2012/0016378; 2012/0046710; 2012/0071949; 2012/0165911; 2012/197375; 2012/0203316; 2012/0203320; 2012/0203321; 2013/0197424; 2013/0197602; 2014/0039587; 2014/0353001; 2014/0358208; 2014/0358209; 2014/0358210; 2015/0045864;
2015/0066120; 2015/0018915; 2015/0051681; U.S. patent application Ser. Nos. 14/557,211 and 14/286,797; and U.S. Provisional Patent Application Ser. No. 62/113,291, all of which are incorporated herein by reference.
Any number of segmented electrodes 130 may be disposed on the lead body 110 including, for example, one, two three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, sixteen or more segmented electrodes 130. It will be understood that any number of segmented electrodes 130 may be disposed along the length of the lead body 110. A segmented electrode 130 typically extends only 75%, 67%, 60%, 50%, 40%, 33%, 25%, 20%, 17%, 15%, or less around the circumference of the lead.
The segmented electrodes 130 may be grouped into sets of segmented electrodes, where each set is disposed around a circumference of the lead 100 at a particular longitudinal portion of the lead 100. The lead 100 may have any number segmented electrodes 130 in a given set of segmented electrodes. The lead 100 may have one, two, three, four, five, six, seven, eight, or more segmented electrodes 130 in a given set. In at least some embodiments, each set of segmented electrodes 130 of the lead 100 contains the same number of segmented electrodes 130. The segmented electrodes 130 disposed on the lead 100 may include a different number of electrodes than at least one other set of segmented electrodes 130 disposed on the lead 100.
The segmented electrodes 130 may vary in size and shape. In some embodiments, the segmented electrodes 130 are all of the same size, shape, diameter, width or area or any combination thereof. In some embodiments, the segmented electrodes 130 of each circumferential set (or even all segmented electrodes disposed on the lead 100) may be identical in size and shape.
Each set of segmented electrodes 130 may be disposed around the circumference of the lead body 110 to form a substantially cylindrical shape around the lead body 110. The spacing between individual electrodes of a given set of the segmented electrodes may be the same, or different from, the spacing between individual electrodes of another set of segmented electrodes on the lead 100. In at least some embodiments, equal spaces, gaps or cutouts are disposed between each segmented electrode 130 around the circumference of the lead body 110. In other embodiments, the spaces, gaps or cutouts between the segmented electrodes 130 may differ in size or shape. In other embodiments, the spaces, gaps, or cutouts between segmented electrodes 130 may be uniform for a particular set of the segmented electrodes 130, or for all sets of the segmented electrodes 130. The sets of segmented electrodes 130 may be positioned in irregular or regular intervals along a length the lead body 110.
Conductor wires (not shown) that attach to the ring electrodes 120 or segmented electrodes 130 extend along the lead body 110. These conductor wires may extend through the material of the lead 100 or along one or more lumens defined by the lead 100, or both. The conductor wires couple the electrodes 120, 130 to the terminals 145.
The sets of segmented electrodes 330 each include either two (
When the lead 100 includes both ring electrodes 120 and segmented electrodes 130, the ring electrodes 120 and the segmented electrodes 130 may be arranged in any suitable configuration. For example, when the lead 100 includes two ring electrodes 120 and two sets of segmented electrodes 130, the ring electrodes 120 can flank the two sets of segmented electrodes 130 (see e.g.,
By varying the location of the segmented electrodes 130, different coverage of the target neurons may be selected. For example, the electrode arrangement of
Any combination of ring electrodes 120 and segmented electrodes 130 may be disposed on the lead 100. For example, the lead may include a first ring electrode 120, two sets of segmented electrodes; each set formed of four segmented electrodes 130, and a final ring electrode 120 at the end of the lead. This configuration may simply be referred to as a 1-4-4-1 configuration (
Any other suitable arrangements of segmented and/or ring electrodes can be used including, but not limited to, those disclosed in U.S. Provisional Patent Application Ser. No. 62/113,291 and U.S. Patent Applications Publication Nos. 2012/0197375 and 2015/0045864, all of which are incorporated herein by reference. As an example, arrangements in which segmented electrodes are arranged helically with respect to each other. One embodiment includes a double helix.
One or more electrical stimulation leads can be implanted in the body of a patient (for example, in the brain or spinal cord of the patient) and used to stimulate surrounding tissue. The lead(s) are coupled to the implantable pulse generator (such as IPG 14 in
In at least some instances, the clinician may be remote from the patient. For example, the clinician may be in another room, treatment or care facility, city, state, or even country. This may be advantageous as it can allow skilled clinicians to interact with patients that are remote from the clinician without requiring travel or loss of time by the clinician. As another example, the patient may be sent home after surgery and the clinician can program the device remotely while the patient is at home.
Such remote programming, however, may encounter difficulties such as, for example, limited bandwidth, speech impaired patients, deaf patients, patients who speak a different language from the clinician, noise over the phone or video connection between patient and clinician, patient difficulty hearing due to hearing aid devices, and the like.
In addition, current techniques for programming an electrical stimulation system (such as by programming IPG 14 in
Accordingly, in at least some instances, it may be desirable to program the implantable pulse generator (or other deep brain stimulation device) using a semi- or wholly automated system such as that described herein that obtains feedback from the patient, generates new stimulator parameter values using machine learning techniques, and tests them to determine whether the parameter values have caused a therapeutic response that is more desirable. This process iterates until stimulation parameter values are determined that cause therapeutic responses within a designated tolerance or until a stop condition is reached. The feedback that is obtained is used in conjunction with the machine learning techniques to explore the stimulation parameter space for parameter values that potentially provide better outcomes for the patient, faster programming of the device, increased battery life, and/or control multiple independent current sources and directional leads. Determining which parameter values cause therapeutic responses within the designated tolerance or within other potentially specifiable parameters (such as with least battery usage and the soonest) is also referred to as finding “optimal” stimulation parameter values.
In some instances, the feedback loop stimulation parameter control system is completely automatic and responses to changes in the stimulation parameter values are obtained from one or more sensors associated with the patient. Having a sensor allows the response to be directly monitored by the feedback loop stimulation parameter control system instead of depending upon potentially subjective entry of patient feedback. This automated system provides a feedback loop to a machine learning based control system that tries to intelligently choose new parameter values based upon the observables from the patient (or other patient data) and a set of rules. Feedback loop programming of a deep brain stimulator using machine learning techniques is discussed in further detail with respect to
In at least one embodiment, a feedback loop stimulation parameter control system such as that described in
In at least some instances, a treating physician may wish to tailor the stimulation parameters (such as which one or more of the stimulating electrode contacts to use, the stimulation pulse amplitude (such as current or voltage amplitude depending on the stimulator being used) the stimulation pulse width, the stimulation frequency, or the like, (or any combination thereof) for a particular patient to improve the effectiveness of the therapy.
The programming procedure performed in conventional electrical stimulation (such as deep brain or spinal cord stimulation) is often performed in an initial session and, in at least some instances, at later sessions. The procedure can involve, for example, testing different sets of stimulation parameters (which can include variations in the electrodes that are selected as well as different electrical parameters such as amplitude, duration, pulse frequency, and the like) and annotating when there is a beneficial therapeutic effect or an unwanted side effect. In at least some embodiments, the clinician performs a monopolar review testing each electrode individually and recording therapeutic/beneficial effects and side effects for each electrode on the lead corresponding to different values of the stimulation amplitude or other stimulation parameters. The clinician may also perform bipolar or multipolar reviews using two or more electrodes.
In contrast to these conventional methods, the feedback loop stimulation parameter control system described here can be used to provide information (e.g., which stimulation parameters to set and to what values) or can be integrated into other components of a stimulation system to set stimulation parameters while programming the device. For example, the feedback loop stimulation parameter control system can be integrated with programming software (such as the IPG software). In addition, the feedback loop stimulation parameter control system can be integrated into clinical response software to provide recommendations or new settings based upon observed (or detected) clinical responses. In at least some embodiments, the stimulation parameter feedback loop control system, programming software, and clinical response software are all integrated.
In at least some of these scenarios, the clinician may use the feedback loop stimulation parameter control system to monitor a patient during programming, to transfer data (such as stimulation parameter values) to another device, or may not be involved—the feedback loop stimulation parameter control system may be used to determine stimulation parameters and set the patient's device without the presence of the clinician.
All such permutations of location of the feedback loop stimulation parameter control system into the general environment of a stimulation system and the extent of involvement of the clinician are contemplated.
The computing device 400 can be a computer, tablet, mobile device, or any other suitable device for processing information. The computing device 400 can be local to the user or can include components that are non-local to the computer including one or both of the processor 402 or memory 404 (or portions thereof). For example, in some embodiments, the user may operate a terminal that is connected to a non-local processor or memory.
Other examples of computing devices can include a watch, wristband, smart phone, or the like. Such computing devices can wirelessly communicate with the other components of the electrical stimulation system, such as the clinician programmer 18 (
The computing device 400 can utilize any suitable processor 402 including one or more hardware processors that may be local to the user or non-local to the user or other components of the computing device. The processor 402 is configured to execute instructions provided to the processor.
Any suitable memory 404 can be used for the computing device 402. The memory 404 illustrates a type of computer-readable media, namely computer-readable storage media. Computer-readable storage media may include, but are not limited to, nonvolatile, non-transitory, removable, and non-removable media implemented in any method or technology for storage of information, such as computer readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (“DVD”) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computing device.
Communication methods provide another type of computer readable media; namely communication media. Communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, data signal, or other transport mechanism and include any information delivery media. The terms “modulated data signal,” and “carrier-wave signal” includes a signal that has one or more of its characteristics set or changed in such a manner as to encode information, instructions, data, and the like, in the signal. By way of example, communication media includes wired media such as twisted pair, coaxial cable, fiber optics, wave guides, and other wired media and wireless media such as acoustic, RF, infrared, Bluetooth™, near field communication, and other wireless media.
The display 406 can be any suitable display or presentation device, such as a monitor, screen, display, or the like, and can include a printer. The input device 408 can be, for example, a keyboard, mouse, touch screen, track ball, joystick, voice recognition system, or any combination thereof, or the like. Another input device 408 can be a camera from which the clinician can observe the patient. Yet another input device 408 is a microphone where the patient or clinician can provide responses or queries.
The electrical stimulation system 412 can include, for example, any of the components illustrated in
In some embodiments, the clinical response values 508 may be weighted according to the time at which the test took place. It has been found that, for at least some patients or at least some testing procedures, patient fatigue occurs over time which may degrade or otherwise change the clinical response values (whether obtained from the patient, clinician, or a sensor). For example, more recent clinical response values 508 may be weighted higher than earlier clinical response values as the more recent clinical response values are more predictive of the clinical response to the adjusted stimulation parameters because these later tests have a similar amount of patient fatigue. Examples of weights can be, but are not limited to, values in a range from 1.05 to 1.5.
Note that in at least some embodiments, the FLSPCS 500 may be executed on its own and is not connected to a controller. In such instances it may be used to merely determine and suggest programming parameters, visualize a parameter space, test potential parameters, etc.
In at least some embodiments, the machine learning engines 502 facilitate the feedback loop stimulation parameter control system (or a user of the system) to explore the possible stimulation parameter space in order to choose values for programming the deep brain simulator controller 505. Any type of machine learning engine, supervised or unsupervised, can be incorporated into the feedback loop stimulation parameter control system 500, including for example, Naïve Bayes classifiers, support vector machines (SVMs), ensemble classifiers, neural networks, Kalman filters, regression analyzers, etc. The role of the machine learning engines 502 in the FLSPCS 500 is to predict the outcomes for various stimulation parameter values to aid (e.g., a clinician) in exploring the stimulation parameter space more effectively and more efficiently to produce results that are optimal, desired, or acceptable. That is, a machine learning engine 502 generally trains itself on data to generate a model of the data. (In supervised machine learning the data is known a priori; in unsupervised machine learning, the model is derived from the data without a known initial structure.) Using the derived data model, the engines 502 can predict information about or classify a new piece of unknown data based upon what it already “knows.”
In this instance, the machine learning engines 502 build models which are used to predict outcomes of sets of stimulation parameter values. A set of stimulation parameter values defines a value for each stimulation parameter being determined. For example, if current amplitude and contact location are the stimulation parameters being explored or adjusted, then a set of stimulation parameter values is a group of values—one for amplitude and one or more that describe the contact location (segmented contacts may have x, y, z directional components as well as a position on the axis of the lead). Outcomes of each set of stimulation parameter values may be further described as “optimal” if it meets certain criteria. Optimal outcome metrics may be predesignated or settable (variable) and refer to attributes such as: 1) time to outcome—fastest optimal outcome (assuming multiple outcomes with same clinical response); 2) best outcome—best possible clinical outcome within the parameter set chosen (global versus local optimal setting); and/or 3) battery savings—optimal outcome with least battery usage (assuming multiple outcomes with same clinical response).
In some embodiments, the machine learning engines 502 can utilize imaging data to inform the choice of the next set of values. This will be particularly useful, when the algorithm finds itself in a region of parameter space that is fairly flat and the choice of next step is not apparent from the patient response alone. Imaging data that provides information about the location of the lead in the patient's brain along with priors informing the algorithm of which directions may be better choices for the next step could lead to faster convergence.
In some embodiments, the machine learning engines 502 determine expected outcomes for parameter values that have not yet been tested based upon what the engines have “learned” thus far and provide a recommendation for a next set of values to test. Here, testing refers to the iterative testing required to find optimal parameters for configuring the deep brain stimulation controller 505. The recommendation for a next set of values to test is based upon which of the determined expected outcomes meet a set of designated (determined, selected, preselected, etc.) criteria (e.g., rules, heuristics, factors, and the like). For example, rules considered may include such factors as: the next set of values cannot be one of the last 10 settings tested or cannot be too close to previously tested setting. Examples of additional such rules are described with reference to
The process for determining expected outcomes for (sets of) parameter values that have not yet been tested may involve use of other data for machine learning. For example, data from other programming sessions for the same patient as well as from other patients may be used to train the machine learning engines 502. In some embodiments, no prior data may be used. In this case, the machine learning engines 502 will use data learned from this patient only in one particular setting. In other embodiments, data from the same patient but from previous sessions may be used. In some embodiments all patient data from all sessions may be used. In some embodiments all patient data utilizing lead location information (knowledge of lead location in space relative to anatomy) may be used. Different other combinations are also possible.
In order to use this data for machine learning purposes, the data is first cleansed, optionally transformed, and then modeled. In some embodiments, new variables are derived, such as for use with directional leads, including central point of stimulation, maximum radius, spread of stimulation field, or the like Data cleansing and transformation techniques such as missing data imputation and dimension reduction may be employed to ready the data for modeling. When there is no prior data used, this step may not be necessary. (The model is built up as more data is observed by the system 500.)
In some embodiments, different types of models may be used to fit stimulation parameter data such as (including but not limited to):
-
- 1. Function-based models (splines, wavelets, kernel regression, additive models, generalized additive models, etc.);
- 2. Spatial or spatial-temporal models (variogram, dynamic or static models, random fields, copulas, etc.);
- 3. Time series models (autoregressive, moving average, autoregressive conditional heteroscedasticity models along with their variants—ARMA, VARMA, VARMAX, ARIMA, GARCH, etc.);
- 4. Latent variable models (factor analysis, structural equation modeling, hidden Markov models, etc.);
- 5. Graph models (Directed Acyclic Graphs, Bayesian networks, undirected graphs, etc.);
- 6. Other models (linear or non-linear regression, mixture models, hierarchical models, etc.)
In addition, each machine learning engine 502, when more than one is used, may derive its own model using types that are distinct from each other (although not necessary). When this occurs, the logic 501 may let each machine learning engine 502 come to its own conclusion for recommending the next set of stimulation parameter values to test. In some instances, the machine learning engines 502 may “vote” on the recommendation with recommendations by engines that have historically provided optimal answers potentially weighted higher than those determined by engines that have not. Other variations and other voting schemes are possible.
The machine learning engines 502 also may desire to determine how best a predicted outcome meets the optimal outcome metrics. To do this, in some embodiments a variety of different optimization techniques may be used, including:
-
- 1. All of the optimization algorithms and estimation procedures that were used to fit the model to the data (e.g., gradient descent, Kalman filter, Markov chain, Monte Carlo, and the like);
- 2. Optimization algorithms reformulated for search (e.g., simulated annealing);
- 3. Spatial interpolation (e.g., kriging, inverse distance weighting, natural neighbor, etc.);
- 4. Supplementary methods that aid the optimization process (e.g., variable selections, regularization, cross validation, etc.); and
- 5. Search algorithms (e.g., “golden section” search, binary search, etc.).
Using these techniques, the machine learning engines 502 can decide whether a particular predicted outcome for a set of stimulation parameter values is the fastest optimal outcome, the best possible clinical outcome, or the optimal outcome with least battery usage, for example.
As mentioned, in some embodiments, a golden section search may be used as an optimization algorithm to determine whether a predicted outcome meets the optimal outcome metrics. The golden section search is performed as follows:
-
- First search in one dimension, then in a subsequent dimension;
- Search dimension is broken into parts, relative clinical score determines an excluded portion of space, and search continues iteratively;
- Search range and/or epsilon value can be adaptive as nearing ideal location;
- For cylindrical lead, dimensions are typically amplitude and contact location; for directional lead, rotation angle may be added.
Other embodiments may support other types of search and/or optimization algorithms. In some embodiments, optimization of results may be performed at the same time as initial search for acceptable outcomes, or may be performed at a different time.
As mentioned, the pulse generator feedback control logic 501 may be used to search and configure (and/or visualize and/or test) different types of stimulation parameters of the various leads potentially causing different effects upon the patient 506 in which the IPG is implanted. Embodiments of a feedback loop stimulation parameter control system 500 may support a variety of different stimulation parameters to be configured, including one or more of the following:
-
- 1. Contact Configurations/Polarities
- a. Monopoles only—only use single contact cathodes
- b. Monopoles and fractionalization—use single contact cathodes or two adjacent contact cathodes (e.g., which electrodes are on/off to steer current)
- c. Bipoles only—use adjacent contacts as anode-cathode, search all possible pairings
- d. Any possible settings—all configurations possible with given system including anodes
- 2. Pulse Widths
- a. Single pulse width—only use a single pulse width for entire search (e.g., 60 us, 30 us)
- b. Multiple pulse widths—allow search to take place between varying pulse widths
- 3. Frequencies
- a. Single frequency—only use a single frequency for entire search (e.g, 130 Hz)
- b. Multiple frequencies—allow search to take place between varying frequencies
- 4. Amplitude
- a. Single amplitude—only use a single amplitude for entire search
- b. Multiple amplitudes—allow search to take place at varying amplitudes
- 1. Contact Configurations/Polarities
Given these possible stimulation parameters, the FLSPCS 500 can move about the parameter space in different orders, by different increments, and limited to specific ranges. Namely, in some embodiments of the FLSPCS 500, techniques for determining incremental steps to test may include one or more of (including in combination where the combination makes sense):
-
- 1. Basic monopolar review—progress along a single contact until side effects are prohibitive, then proceed to adjacent control
- 2. Steering—select contact, increase amplitude until therapeutic effect is seen, steer along main axis of the lead (move up and down and around—circumnavigate—the lead)
- 3. Model based—select next step according to parameters selected by the model, based on previous parameters or previous patients' experience
- 4. Random—select next step at random; may be among entire space or among a subset of candidates (may be used at beginning of session to explore the space, then move to one of the other techniques)
- 5. Spatial search algorithm—use optimized search algorithm to traverse stimulation parameter space and find optimal settings.
In some embodiments, step size limitations may include:
-
- 1. No limit—can test any parameter combination
- 2. Step size limit based off of previous step
- 3. Step size limit based off of all previous search points
- 4. Step size limit based off of the number of steps already taken.
In some embodiments, search range limitations may include one or more of (including in combination where makes sense):
-
- 1. No limits—can test any parameter combination
- 2. Amplitude limit—can only test to a certain amplitude
- 3. Charge limit—can only test to certain amplitude*pulse width limit
- 4. Defined limit—physician/clinician does a monopolar review-like search to identify side effect boundaries at each contact; these are then used as inputs for search bounds.
Different combinations of increments, increment sizes, search range limitations or other parameters affecting movement about the parameter space may be incorporated.
In addition, the feedback loop stimulation parameter control system 500 may allow the user to provide search range limitations to one or more of the stimulation parameters to limit the range for that stimulation parameter over which the system will search for parameters. For example, the user may restrict which electrodes can be used for stimulation or may restrict the amplitude or pulse width to a certain range or with a selected maximum or minimum. As one illustration, based on the site of implantation, the user may be aware that the distal-most and proximal-most electrodes are unlikely to produce suitable stimulation and the user limits the range of electrodes to exclude these two electrodes.
For a lead with segmented electrodes, the number of possibilities for parameter selection can be very large when combinations of electrodes and different amplitudes on each electrode are possible. In some embodiments using a lead with segmented electrodes, the selection of electrodes used for stimulation may be limited to fully directional selections (i.e., selection of only a single segmented electrode) and fully concentric selections (i.e., all electrodes in a single set of segmented electrodes are active with the same amplitude). In other embodiments, the initial movement through parameter space may be limited to fully directional and fully concentric selections. After a set of stimulation parameters is identified using these limits, variation in the selection of electrodes may be opened up to other possibilities near the selection in the identified set of stimulation parameters to further optimize the stimulation parameters.
In some embodiments, the number of stimulation parameters that are varied and the range of those variations may be limited. For example, some stimulation parameters (e.g., electrode selection, amplitude, and pulse width) may have larger effects when varied than other stimulation parameters (e.g., pulse shape or pulse duration). The movement through parameter space may be limited to those stimulation parameters which exhibit larger effects. In some embodiments, as the feedback loop stimulation parameter control system 500 proceeds through testing of sets of stimulation parameters, the system may observe which stimulation parameters provide larger effects when varied and focus on exploring variation in those stimulation parameters.
In at least some embodiments, the feedback loop stimulation parameter control system 500 includes a user interface (not shown in
In some embodiments where a deep brain stimulator is actually configured via a controller 505 with at least one set of stimulation parameter values forwarded by pulse generator feedback control logic 501, the physician/clinician may monitor the patient throughout the process and record clinical observables in addition to the patient 506 being able to report side effects. When a side effect is observed/recorded, the various search algorithms will take that fact into account when selecting/suggesting a next set of values to test. In some embodiments, for example, those that select contacts via monopolar review, other parameters may be changed until they cause a side effect, which case is noted as a boundary. For example, in monopolar review where amplitude is another stimulation parameter being varied, the amplitude may be increased progressively until a side-effect is observed.
In some embodiments, more than one clinical metric (e.g., tremor, bradykinesia, etc.) may be important observables. Different embodiments of the FLSPCS 500 may handle these metrics differently. For example, some embodiments might identify an ideal location for each metric and choose one ideal location between them, set in the patient's remote controller so the patient can choose as needed, or chose a best combined outcome. As another example, some embodiments may search multiple outcomes at the same time and use the best combined score as the best outcome or find a best location for each metric individually. As yet another example, some embodiments may use a sequential process for selecting stimulation parameter values for multiple outcomes. For example, a system may search parameter space for a first outcome (e.g., bradykinesia) and, upon finding a suitable end condition, then search parameter space for a second outcome (e.g., rigidity). While searching parameter space for the first outcome, clinical response values for both the first and second outcomes can be obtained. Thus, when the system switches to the second outcome there are already a number of clinical response values for that outcome which will likely reduce the length of the search.
In some embodiments, two stimulation leads may be implanted to produce stimulation effects on two sides of the body (e.g., the right and left sides of the body). The same procedure described herein can be used to either jointly determine the stimulation parameters for the two leads by exploring the joint parameter space or individually determine stimulation parameters for the two leads by exploring the parameter space for each lead individually. In some embodiments, the user may determine for each side of the body which clinical response is dominant or most responsive. This may be done, for example, by having the patient perform a single task which captures multiple responses (e.g. connecting dots on the screen to monitor tremor and bradykinesia of the movement) or a small series of tasks. This enables the system to determine which clinical response to use to identify the stimulation parameters for that side of the body. #
As noted, the feedback may be provided directly by the patient 506, entered by an observer such as a clinician (not shown), or may be provided by means of a sensor 507 associated with and in physical, auditory, or visual contact with the patient 506. In an embodiment where the feedback can be monitored automatically or semi-automatically, such as with use of sensor 507, it may not be necessary for a clinician or other observer to be present to operate the FLSPCS 500. Accordingly, in such embodiments a user interface may not be present in system 500.
More specifically, in
In
In
In
In
In some embodiments, the feedback loop stimulation parameter control system may suggest to the user the amount of wash-in time to wait after applying stimulation at the particular stimulation settings before obtaining a clinical response value. In at least some cases, it is found that when stimulating at new parameters which are close in proximity to the stimulation parameters which immediately preceded them (e.g. stimulating using an electrode at 2 mA, then stimulating using the same electrode at 2.5 mA), that there does not need to be as long of a wash-in time as when the parameters are far apart. Based on the distance from the previous stimulation parameters, the system can suggest an appropriate wash-in time. Alternatively or additionally, it may be desirable to test the patient's specific wash-in time for the specific clinical response being evaluated. For instance, at the start of a programming session, the patient would be programmed with some generic, acceptable stimulation parameters and we would measure the clinical response every 30 seconds and then observing how long after initially changing the settings it would take for the response to asymptote and come to a steady response. That would be used as the wash-in time for that patient or hemisphere.
In
In
In
In
As described with reference to
-
- 1. The centroid of the stimulation (x, y, z coordinates) plus amplitude.
- 2. The centroid of the stimulation (x, y, z coordinates) plus single value to describe the radius.
- 3. The centroid of the stimulation (x, y, z coordinates) plus three values to describe the radii (also x, y, z).
- 4. The centroid of the stimulation (x, y, z coordinates) plus volume.
- 5. The centroid of the stimulation (x, y, z coordinates) plus two values to describe cross-sectional surface areas (e.g. x-y and y-z).
- 6. Pre-computing the volumes for each of the steering states that are created by our programming controls. Typically, the volumes generated are coherent volumes which are approximately spherical. These generated volumes could then be searched for the volume that is most closely aligned with the desired volume and the corresponding stimulation setting identified.
- 7. Using the three cylindrical coordinates (z, theta, r) to approximate the longitudinal position, rotational position, and amplitude (ignoring spread control).
- 8. Current fractionalizations assigned to each contact, plus overall amplitude (this would be 9 dimensions, and would likely require a method such as machine learning to handle).
- 9. Dividing the process up into several stages, where “Ring Mode” is first used to search for best longitudinal position and amplitude, then best rotational position and amplitude, and then best spread and amplitude. Ring Mode refers to stimulation settings where the electrodes on any active directional row of a lead can be treated like a cylindrical electrode by activating them at approximately the same percentage (e.g., 33% for a row with three electrodes).
In block 702, the logic generates an initial proposed set of stimulation parameter (SP) values. These include a value for each stimulation parameter being adjusted. For example, when the set of stimulation parameters include contact location and amplitude, the initial proposed set may be “contact=3” and “amplitude=2 mA” as shown in
In one example embodiment, the initial proposed set of stimulation parameter values is chosen arbitrarily because data has not been collected yet for this patient. In another example embodiment, the logic is expanded to use the distribution of optimal settings (e.g. amplitude and/or contact location) from previous patients to pick the starting setting for a new patient. This allows clinicians to begin the programming session with settings that other patients have found beneficial. In addition, new patients may be matched to previous patients via external information, such as diagnosis, imaging data, or baseline symptom severity.
In block 703, the logic forwards the initial proposed set of stimulation parameter values to the deep brain stimulator. In one embodiment, this is done by forwarding the parameter values to the DBS controller (see message 504 in
In blocks 704-707, the logic performs a loop (iterates) to received therapeutic responses, predict and propose new stimulation parameter values to try, and forwards them to set the DBS, until a stop condition is reached or a clinical response value that corresponds to a therapeutic response that is within a designated threshold is reached.
Specifically, in block 704, the logic determines whether an end condition is reached. For example, the logic may determine whether a clinical response has been received that corresponds to a therapeutic response within a designated threshold. For example, if tremor is what is being observed, a (beneficial) therapeutic response that is under 2.0 (on a scale of 0 to 4) may be considered a sufficient stopping point. In some embodiments, other stop conditions are designated such as a certain number of iterations has been performed, the therapeutic response changes less than x % with y number of iterations, or the user is satisfied with the result. Other stop conditions are possible.
In block 705, the logic receives indications of clinical response values. As discussed above, these values may be received from a patient by observation, or directly from a sensor associated with the patient, for example, an accelerometer located on the patient's wrist. In some embodiments, the indications of clinical values may be received from a clinician or physician observing the patient.
In block 706, the logic executes additional logic steps to generate a revised proposed set of stimulation parameter values incorporating machine learning techniques. Examples of these additional steps are described in further detail with reference to
In block 707, the logic forwards the revised proposed set of stimulation parameter values to set the DBS to the revised values and returns to the beginning of the loop in block 704.
In some embodiments, steps 704 through 707 are performed for a first therapeutic response while obtaining clinical response values for both the first therapeutic response and a second therapeutic response. After finding the end condition for the first therapeutic response, steps 704 through 707 can be performed for a second therapeutic response. This procedure can be repeated for other therapeutic responses and provides a sequential process for selecting stimulation parameter values to obtain multiple therapeutic effects. For example, the first therapeutic effect may be bradykinesia and the second therapeutic effect can be rigidity. Steps 704 through 707 are preformed until an end condition is found for bradykinesia, although clinical response values are obtained for both bradykinesia and rigidity. Thus, when steps 704 through 707 are subsequently performed for rigidity, there are already a number of clinical response values existing for rigidity.
In addition, these techniques described by
Specifically, in block 801, the logic automatically generates, using machine learning techniques, one or more therapeutic response values for stimulation parameter values that have not yet been tested. For example, in an embodiment in which the set of stimulation parameters include two parameters: contact location and amplitude, it may be desirable to visualize (and set) the values for contact location from 1 to 8 and amplitude from 0 mA to 6 mA in 0.1 increments. If one were to visualize these possible values on a clinical effects map (like clinical effects map 620 in
At programming step t+1 (e.g., iteration of the loop shown in
For example, if m=2 (contact location and amplitude), then “X,” the t×m matrix, may be represented as a sparse matrix 901 illustrated in
A number of machine learning methods can accomplish this task as described earlier. If there are two or more adverse symptoms to control, then the prediction phase may generate a separate prediction for each therapeutic response for each matrix location. Alternatively, the prediction phase may generate a combined score based upon combining the multiple systems in a variety of ways, including weighting them based upon importance, aggregating them, and the like.
In some embodiments, multiple machine learning methods may be used (concurrently) to generate multiple predictions and the results scored in some manner. For example, when programming deep brain stimulators, patient response to stimulation is unlikely to follow an established parametric distribution, is prone to noise, and is patient-specific. In some embodiments, predictions from several machine learning methods can be stored in memory and then analyzed for accuracy after new data are collected. Machine learning methods that produce accurate predictions will be “up-voted” while methods that produce inaccurate predictions will be “down-voted”. This process allows high performing prediction methods to drive the search process, as a single method may be unlikely to work for all patients or for the entire duration of a programming session.
The prediction techniques of block 801 could also take into account previously acquired local field potential (LFP) data, which would give an approximation of the prediction space.
In block 802, the logic optionally displays or presents the predictions resulting from block 801. As noted in
In at least some embodiments, a confidence score can be generated and displayed to indicate the likelihood that the predictions will produce a good outcome or the predicted outcome. For example, the confidence score may be based on repeatability of the responses (e.g. if the same or similar settings are used or predicted and the responses vary too much, the outcome may be questionable), or the confidence score may be based on robustness of the clinical response (e.g. if all of the patient responses are a 3.9 or 4.0 out of 4, then the procedure will likely not converge to a good solution).
In block 803, the logic performs phase two of the process: it automatically proposes a next set of stimulation parameters to test. In an example embodiment, this phase takes into account one or more of the predictions of therapeutic responses for parameter values that have not yet been tested and may also incorporate rules and/or heuristics to provide more efficient selection of a next proposed set of stimulation parameter values. In some embodiments, the rules/heuristics are directed to eliminating those values that are unlikely to yield desirable results for one reason or another. For example, in one embodiment, inappropriate grid locations are eliminated using a set of rules, then the R remaining locations are ranked in different ways according to predefined criteria, for example from 1=“best” to R=“worst.” In this case, the grid location chosen for the next programming step is the location with the lowest overall rank. In some embodiments, an option is provided where several of the top candidates are returned, thereby giving the programmer/clinician more manual control over the programming path. In addition, an option to completely override the suggestion(s) provided by the logic is also possible.
In one embodiment, the following example rules are incorporated to eliminate inappropriate sets of stimulation parameter values (e.g. locations in the grid of 4331 values):
-
- 1. Cannot be one of the last 10 settings (values) tested, i.e. re-testing is possible though unlikely.
- 2. Cannot be too close to previously tested settings, where closeness (or mathematical proximity) is defined by a mathematical function dependent on the number of programming steps already taken.
- 3. Cannot have an amplitude that is 1.5 mA higher than the last setting tested.
- 4. If an adverse side effect is observed in previous programming steps, then all settings of higher amplitude and within ±0.3 of the contact location of the setting which caused the side effect are eliminated.
In some embodiments, the mathematical function used to determine closeness to previous determined settings may take into account a “geometry” associated with the stimulation parameter values formed by already viewed grid locations. For example, a next set of stimulation parameter values may be located in the center of this geometry.
Other embodiments may incorporate additional or different rules.
Next, the remaining possible sets of stimulation parameter values (e.g. remaining grid locations) are ranked. For example, in one embodiments, the sets of values are given three rankings according to the following criteria:
-
- 1. Rank by predicted therapeutic response, from best to worst.
- 2. Rank by total distance to already tested locations, from farthest to closest.
- 3. Rank by similarity in distances to already tested locations, from most similar to least similar.
The relative importance of these rankings may shift during a DBS programming session. In the beginning of the session, the goal is to explore the parameter space, so distance rankings are more important. Towards the end of a session, it may be necessary to pinpoint an optimal grid location, so the therapeutic response ranking may become more important. A mathematical function can be used to summarize the three rankings into one overall ranking according to how far the programming session has progressed. In other words, there may be a gradual decrease in the acceptance of a worse solution (unfavorable predicted therapeutic response) to allow for early exploration of the parameter space.
The logic can gradually be made more complex by adding more rankings. If there are two adverse symptoms to control, then the proposal phase can be adapted by adding a fourth ranking based on predictions for the second therapeutic response. In addition, in some embodiments an option can be added to allow external inputs to lock in the importance of the therapeutic response rankings, for scenarios where a clinician wishes to emphasize a particular response. Multiple predictions from different machine learning methods can also be incorporated into the proposal phase by adding more rankings. For each new feature added, fine-tuning of how the ranks come together and vary in importance during the programming session is specified.
When the number of stimulation parameters in a set is greater than 2, as is the case for directional leads, an exhaustive search through all the possible parameter sets (e.g., grid locations) becomes prohibitively slow. In one embodiment, the logic adapted such that only a subset of possible sets of stimulation parameter values (e.g., grid locations) is examined at each programming step. For example, in one example directional lead algorithm, the starting point is to determine an optimal grid location for the omnidirectional ring mode, which is equivalent to finding an optimal grid location in a 4-contact conventional lead. After the parameter space has been narrowed down to the vicinity of the best ring mode configuration, a 3-dimensional space around the lead can be constructed. The logic can then proceed to search in m=4 dimensions (x, y, z coordinates plus amplitude) in the same way that it searches in m=2 dimensions. A separate algorithm can convert the 3-dimensional location to stimulation settings.
Recognizing that the control logic can learn from previous programming sessions (for this patient or other patients, the logic can steer the search path towards parameter regions with a high density of prior optimal settings. By testing settings that have worked for other patients first, the clinician will likely need less programming steps to optimize therapeutic response. This logic may be incorporated into rules for proposing a next set of stimulation parameter values.
In block 806, the logic returns the proposed set of stimulation parameter values. As described with reference to
It will be understood that the system can include one or more of the methods described hereinabove with respect to
It will be understood that each block of the flow diagram illustrations, and combinations of blocks in the flow diagram illustrations and methods disclosed herein, can be implemented by computer program instructions. These program instructions or code logic may be provided to a processor to produce a machine, such that the instructions, which execute on the processor, create means for implementing the actions specified in the flow diagram block or blocks disclosed herein. The computer program instructions may be executed by a processor to cause a series of operational steps to be performed by the processor to produce a computer implemented process. The computer program instructions may also cause at least some of the operational steps to be performed in parallel. Moreover, some of the steps may also be performed across more than one processor, such as might arise in a multi-processor computer system. In addition, one or more processes may also be performed concurrently with other processes, or even in a different sequence than illustrated without departing from the scope or spirit of the invention.
The computer program instructions can be stored on any suitable computer-readable medium including, but not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (“DVD”) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computing device.
The above specification provides a description of the structure, manufacture, and use of the invention. Since many embodiments of the invention can be made without departing from the spirit and scope of the invention, the invention also resides in the claims hereinafter appended.
Claims
1. A system for facilitating programming settings of an implantable pulse generator associated with a patient, the system comprising:
- a processor;
- a memory; and
- control logic, stored in the memory, and configured, when executed by the processor, to perform actions comprising:
- a) obtaining a set of stimulation parameter values;
- b) receiving one or more clinical response values arising from testing of the set of stimulation parameter values by stimulating the patient using the set of stimulation parameter values;
- c) predicting one or more therapeutic response values for each of a plurality of sets of untested stimulation parameter values using one or more machine learning engines and the one or more received clinical response values;
- d) selecting, based on at least the one or more predicted therapeutic response values, a revised set of stimulation parameter values from the plurality of sets of untested stimulation parameter values;
- e) receiving one or more clinical response values arising from testing of the revised set of stimulation parameter values by stimulating the patient using the revised set of stimulation parameter values; and
- f) repeating actions c) to e) until a one of at least one stop condition has been reached.
2. The system of claim 1, wherein the selecting comprises selecting, based on at least the one or more predicted therapeutic response values and on a distance of each set of untested stimulation parameter values from one or more tested sets of stimulation parameter values, the revised set of stimulation parameter values from the plurality of sets of untested stimulation parameter values.
3. The system of claim 1, wherein the system further comprises at least one sensor, wherein each instance of receiving one or more clinical response values comprises automatically receiving at least one of the one or more clinical response values from the at least one sensor.
4. The system of claim 1, wherein the at least one stop condition comprises at least one of the following: the one or more received clinical response values indicates a value that corresponds to a therapeutic response indication within a designated tolerance; a predefined number of iterations; a predefined time spent in a programming session; or a determination by a user to stop.
5. The system of claim 1, wherein the predicting comprises using multiple machine learning techniques to predict the one or more therapeutic response values.
6. The system of claim 1, wherein the predicting comprises using multiple machine learning engines to generate multiple predictions of each of the one or more predicted therapeutic response values.
7. The system of claim 1, wherein the selecting comprises selecting the revised set of stimulation parameter values based upon one or more rules.
8. The system of claim 7, wherein the one or more rules includes at least one of: 1) a rule based upon retrying stimulation locations within a number of iterations; 2) a rule based upon mathematical proximity of the untested sets of stimulation parameter values to the sets of stimulation parameter values already tested; 3) a rule based upon values of a stimulation parameter leading to adverse side effects, or 4) a rule based upon a size of a step between values of a same stimulation parameter.
9. The system of claim 1, wherein the selecting comprises selecting the revised set of stimulation parameter values in each repetition of steps c) to e) in contact order, wherein contact order means testing along a first contact on a lead until side effects are prohibitive before testing a next contact on the lead.
10. The system of claim 1, the selecting comprises selecting the revised set of stimulation parameter values in each repetition of steps c) to e) by selectively choosing contacts along a lead and varying other stimulation parameters for each selectively chosen contact.
11. The system of claim 1, wherein t the selecting comprises selecting the revised set of stimulation parameter values in each repetition of steps c) to e) randomly, based upon at least one of: a model of the patient, a spatial search algorithm, or by varying a distance to the revised set of stimulation parameter values.
12. A computer-implemented method for automatically determining patient programming settings for an electrical stimulator, comprising:
- a) obtaining a set of stimulation parameter values;
- b) receiving one or more clinical response values arising from testing of the set of stimulation parameter values by stimulating a patient using the set of stimulation parameter values;
- c) predicting one or more therapeutic response values for each of a plurality of sets of untested stimulation parameter values using one or more machine learning engines and the one or more received clinical response values;
- d) selecting, based on at least the one or more predicted therapeutic response values, a revised set of stimulation parameter values from the plurality of sets of untested stimulation parameter values;
- e) receiving one or more clinical response values arising from testing of the revised set of stimulation parameter values by stimulating the patient using the revised set of stimulation parameter values; and
- f) repeating steps c) to e) until a one of at least one stop condition has been reached.
13. The method of claim 12, wherein the predicting comprises predicting the one or more therapeutic response values using a training data set based upon prior data for one or more subjects.
14. The method of claim 12, wherein the selecting comprises selecting, based on at least the one or more predicted therapeutic response values and on a distance of each set of untested stimulation parameter values from one or more tested sets of stimulation parameter values, the revised set of stimulation parameter values from the plurality of sets of untested stimulation parameter values.
15. The method of claim 12, wherein the at least one stop condition comprises at least one of the following: the one or more received clinical response values indicates a value that corresponds to a therapeutic response indication within a designated tolerance; a predefined number of iterations; a predefined time spent in a programming session; or a determination by a user to stop.
16. The method of claim 12, wherein the predicting comprises using multiple machine learning techniques to predict the one or more therapeutic response values.
17. A computer-readable memory medium containing instructions that control a computer processor, when executed, to determine programming settings of an implantable pulse generator (IPG) associated with a patient by performing a method comprising:
- a) obtaining a set of stimulation parameter values;
- b) receiving one or more clinical response values arising from testing of the set of stimulation parameter values by stimulating the patient using the set of stimulation parameter values;
- c) predicting one or more therapeutic response values for each of a plurality of sets of untested stimulation parameter values using the one or more machine learning engines and the one or more received clinical response values;
- d) selecting, based on at least the one or more predicted therapeutic response values, a revised set of stimulation parameter values from the plurality of sets of untested stimulation parameter values;
- e) receiving one or more clinical response values arising from testing of the revised set of stimulation parameter values by stimulating the patient using the revised set of stimulation parameter values; and
- f) repeating actions c) to e) until a one of at least one stop condition has been reached.
18. The computer-readable memory medium of claim 17, wherein the selecting comprises selecting, based on at least the one or more predicted therapeutic response values and on a distance of each set of untested stimulation parameter values from one or more tested sets of stimulation parameter values, the revised set of stimulation parameter values from the plurality of sets of untested stimulation parameter values.
19. The computer-readable memory medium of claim 17, wherein the predicting comprises using multiple machine learning engines to generate multiple predictions of each of the one or more predicted therapeutic response values.
20. The computer-readable memory medium of claim 17, wherein the predicting comprises using multiple machine learning techniques to predict the one or more therapeutic response values.
Type: Application
Filed: Jul 19, 2023
Publication Date: Nov 9, 2023
Inventors: David Blum (Boston, MA), Sherry Lin (Glendale, CA), Hemant Bokil (Cambridge, MA), Michael A. Moffitt (Solon, OH)
Application Number: 18/223,984
