Abstract: Techniques are disclosed for using feature delineation to reduce the impact of machine learning cardiac arrhythmia detection on power consumption of medical devices. In one example, a medical device performs feature-based delineation of cardiac electrogram data sensed from a patient to obtain cardiac features indicative of an episode of arrhythmia in the patient. The medical device determines whether the cardiac features satisfy threshold criteria for application of a machine learning model for verifying the feature-based delineation of the cardiac electrogram data. In response to determining that the cardiac features satisfy the threshold criteria, the medical device applies the machine learning model to the sensed cardiac electrogram data to verify that the episode of arrhythmia has occurred or determine a classification of the episode of arrhythmia.

Abstract: Techniques that include applying machine learning models to episode data, including a cardiac electrogram, stored by a medical device are disclosed. In some examples, based on the application of one or more machine learning models to the episode data, processing circuitry derives, for each of a plurality of arrhythmia type classifications, class activation data indicating varying likelihoods of the classification over a period of time associated with the episode. The processing circuitry may display a graph of the varying likelihoods of the arrhythmia type classifications over the period of time. In some examples, processing circuitry may use arrhythmia type likelihoods and depolarization likelihoods to identify depolarizations, e.g., QRS complexes, during the episode.

Abstract: Techniques are disclosed for explaining and visualizing an output of a machine learning system that detects cardiac arrhythmia in a patient. In one example, a computing device receives cardiac electrogram data sensed by a medical device. The computing device applies a machine learning model, trained using cardiac electrogram data for a plurality of patients, to the received cardiac electrogram data to determine, based on the machine learning model, that an episode of arrhythmia has occurred in the patient and a level of confidence in the determination that the episode of arrhythmia has occurred in the patient. In response to determining that the level of confidence is greater than a predetermined threshold, the computing device displays, to a user, a portion of the cardiac electrogram data, an indication that the episode of arrhythmia has occurred, and an indication of the level of confidence that the episode of arrhythmia has occurred.

Abstract: Techniques are disclosed for monitoring a patient for the occurrence of a cardiac arrhythmia. A computing system generates sample probability values by applying a machine learning model to sample patient data. The machine learning model determines a respective probability value that indicates a probability that the cardiac arrhythmia occurred during each respective temporal window. The computing system outputs a user interface comprising graphical data based on the sample probability values and receives, via the user interface, an indication of user input to select a probability threshold for a patient. The computing system receives patient data for the patient and applies the machine learning model to the patient data to determine a current probability value. In response to the determination that the current probability exceeds the probability threshold for the patient, the computing system generates an alert indicating the patient has likely experienced the occurrence of the cardiac arrhythmia.

Abstract: Techniques are disclosed for using both feature delineation and machine learning to detect cardiac arrhythmia. A computing device receives cardiac electrogram data of a patient sensed by a medical device. The computing device obtains, via feature-based delineation of the cardiac electrogram data, a first classification of arrhythmia in the patient. The computing device applies a machine learning model to the received cardiac electrogram data to obtain a second classification of arrhythmia in the patient. As one example, the computing device uses the first and second classifications to determine whether an episode of arrhythmia has occurred in the patient. As another example, the computing device uses the second classification to verify the first classification of arrhythmia in the patient. The computing device outputs a report indicating that the episode of arrhythmia has occurred and one or more cardiac features that coincide with the episode of arrhythmia.

Abstract: Techniques are disclosed for monitoring a patient for the occurrence of cardiac arrhythmias. A computing system obtains a cardiac electrogram (EGM) strip for a current patient. Additionally, the computing system may apply a first cardiac rhythm classifier (CRC) with a segment of the cardiac EGM strip as input. The first CRC is trained on training cardiac EGM strips from a first population. The first CRC generates first data regarding an aspect of a cardiac rhythm of the current patient. The computing system may also apply a second CRC with the segment of the cardiac EGM strip as input. The second CRC is trained on training cardiac EGM strips from a smaller, second population. The second CRC generates second data regarding the aspect of the cardiac rhythm of the current patient. The computing system may generate output data based on the first and/or second data.

Abstract: Techniques are disclosed for preparing data for use in artificial intelligence (AI)-based cardiac arrhythmia detection. In accordance with the techniques of this disclosure, a computing system may obtain a cardiac electrogram (EGM) strip that represents a waveform of a cardiac rhythm of a same patient. Additionally, the computing system may preprocess the cardiac EGM strip. The computing system may then apply a deep learning model to the preprocessed cardiac EGM strip to generate arrhythmia data indicating whether the cardiac EGM strip represents one or more occurrences of one or more cardiac arrhythmias.