Abstract: In at least one embodiment, a reinforcement-learning-based searching approach is used to produce a training configuration for a machine-learning model. In at least one embodiment, 3D medical image segmentation is performed using learned image preprocessing parameters.

Abstract: In at least one embodiment, a reinforcement-learning-based searching approach is used to produce a training configuration for a machine-learning model. In at least one embodiment, 3D medical image segmentation is performed using learned image preprocessing parameters.

Abstract: The present invention provides adaptive lightweight acoustic signal classification for physiological monitoring applications. In an exemplary implementation, the total energy of a segment of an acoustic signal recording body sounds is first determined. For each of a plurality of signal classes (e.g., good, noisy, weak), the probability that the segment belongs to the signal class is then calculated using the total energy and profile data for the signal class. The segment is then assigned to one of the plurality of signal classes by reference to the probabilities. Physiological data are then selectively generated and outputted using the segment, depending on the assigned signal class, and the segment is selectively applied as feedback to update profile data for the assigned signal class.

Abstract: A sleep monitoring method and device classify segments of an acoustic physiological signal captured during sleep as snore and apnea segments. The method and device employ a phase-locked loop array to rapidly detect snore segments for widely variant snoring rhythms exhibited by different people or the same person over time. The phase-locked loop array integrates seamlessly with an apnea timer-thresholding mechanism that detects apnea segments.

Abstract: Physiological signal quality classification methods and systems designed to improve ambulatory monitoring. Physiological signals are classified as good, noisy or weak based on signal properties. Once classified, signals are processed differently depending on their classification in order to encourage reliance on reliable physiological data, discourage reliance on unreliable physiological data and induce action to improve signal quality. For example, for a good signal, physiological data may be extracted from the signal and displayed to a person being monitored. For a noisy signal, a noisy signal notification may be displayed to the person in lieu of extracted physiological data. For a weak signal, a weak signal notification may be displayed to the person in lieu of extracted physiological data.

Abstract: A sleep monitoring method and device classify segments of an acoustic physiological signal captured during sleep as snore and apnea segments. The method and device employ a phase-locked loop array to rapidly detect snore segments for widely variant snoring rhythms exhibited by different people or the same person over time. The phase-locked loop array integrates seamlessly with an apnea timer-thresholding mechanism that detects apnea segments.

Abstract: Recursive least squares (RLS) adaptive acoustic signal filtering for a physiological monitoring system reduces residual heart sound in a primary signal remaining after application of a respiration sound bandpass filter to a first instance of a mixed signal containing respiration sound and heart sound. Residual heart sound in the primary signal is reduced by minimizing a component in the primary signal that correlates with a reference signal containing heart sound but almost no residual respiration sound after application of a heart sound bandpass filter to a second instance of the mixed signal. The correlative component in the primary signal is minimized by applying an adaptive filter to the reference signal and subtracting the filtered reference signal from the primary signal to produce a residue signal, wherein the coefficients for the adaptive filter are selected to minimize the least square error of the residue signal.

Abstract: Physiological signal quality classification methods and systems designed to improve ambulatory monitoring. Physiological signals are classified as good, noisy or weak based on signal properties. Once classified, signals are processed differently depending on their classification in order to encourage reliance on reliable physiological data, discourage reliance on unreliable physiological data and induce action to improve signal quality. For example, for a good signal, physiological data may be extracted from the signal and displayed to a person being monitored. For a noisy signal, a noisy signal notification may be displayed to the person in lieu of extracted physiological data. For a weak signal, a weak signal notification may be displayed to the person in lieu of extracted physiological data.

Abstract: A method and device for physiological parameter estimation, such as heart rate estimation, use a phase-locked loop to dynamically track a dominant frequency of an acoustic signal within a frequency band for the physiological parameter and estimate the physiological parameter using the dominant frequency. Because the phase-locked loop starts searching for the current dominant frequency near the most recently identified dominant frequency, the method and device rapidly incorporate slow changes in the dominant frequency that likely reflect real changes in the monitored physiological parameter while being slow to incorporate rapid changes in the dominant frequency that are likely caused by large noise. The hysteresis in phase-locked loop tracking allows the method and device to avoid physiological parameter estimation error induced by short-term large noise and quickly reacquire the dominant frequency once short-term large noise abates.

Abstract: Physiological signal quality classification methods and systems for ambulatory monitoring. Physiological signals are classified as good, noisy or weak based on signal properties. Once classified, signals are processed differently depending on their classification For example, for a good signal, physiological data may be extracted from the signal and displayed to a person being monitored. For a noisy signal, a noisy signal notification may be displayed to the person in lieu of extracted physiological data. For a weak signal, a weak signal notification may be displayed to the person in lieu of extracted physiological data. Moreover, a noisy or weak signal notification displayed to a person being monitored may be accompanied by a corrective action recommendation, such as “move to quieter environment” for a noisy signal or “check body placement of sensor” for a weak signal.

Abstract: A method and device for physiological parameter estimation, such as heart rate estimation, use a phase-locked loop to dynamically track a dominant frequency of an acoustic signal within a frequency band for the physiological parameter and estimate the physiological parameter using the dominant frequency. Because the phase-locked loop starts searching for the current dominant frequency near the most recently identified dominant frequency, the method and device rapidly incorporate slow changes in the dominant frequency that likely reflect real changes in the monitored physiological parameter while being slow to incorporate rapid changes in the dominant frequency that are likely caused by large noise. The hysteresis in phase-locked loop tracking allows the method and device to avoid physiological parameter estimation error induced by short-term large noise and quickly reacquire the dominant frequency once short-term large noise abates.

Abstract: The present invention provides adaptive lightweight acoustic signal classification for physiological monitoring applications. In an exemplary implementation, the total energy of a segment of an acoustic signal recording body sounds is first determined. For each of a plurality of signal classes (e.g., good, noisy, weak), the probability that the segment belongs to the signal class is then calculated using the total energy and profile data for the signal class. The segment is then assigned to one of the plurality of signal classes by reference to the probabilities. Physiological data are then selectively generated and outputted using the segment, depending on the assigned signal class, and the segment is selectively applied as feedback to update profile data for the assigned signal class.

Abstract: Linear classification is used to determine the quality of acoustic physiological signal samples. A feature dataset is extracted from acoustic physiological signal samples of known quality (i.e., weak, noisy, good) acquired over a sampling period. A linear discriminant analysis is performed on the feature dataset to determine a direction of a linear classifier for the feature dataset. A classification error risk analysis is performed on the feature dataset to determine an offset of the linear classifier. The linear classifier is used to classify into reliability classes acoustic physiological signal samples acquired over an operating period. Information is selected for outputting using the assigned classifications, and is outputted.

Abstract: A method and system that reliably estimates a respiration parameter from an acoustic physiological signal without introducing undue complexity or intense computation. A median filter is applied to an energy envelope of the signal to remove heart sound “sparks” from the envelope and better isolate lung sounds. The median filter is followed by a low-pass filter that removes abrupt changes in the envelope caused by the median filter's nonlinearity. Various peak cross-checks are performed on an autocorrelation result generated from the envelope to confirm the reliability of the signal before an estimate of a respiration parameter is generated from the autocorrelation result.

Abstract: Physiological signal quality classification methods and systems designed to improve ambulatory monitoring. Physiological signals are classified as good, noisy or weak based on signal properties. Once classified, signals are processed differently depending on their classification in order to encourage reliance on reliable physiological data, discourage reliance on unreliable physiological data and induce action to improve signal quality. For example, for a good signal, physiological data may be extracted from the signal and displayed to a person being monitored. For a noisy signal, a noisy signal notification may be displayed to the person in lieu of extracted physiological data. For a weak signal, a weak signal notification may be displayed to the person in lieu of extracted physiological data.