Systems and methods to detect a false alarm in a patient monitoring device

A method, comprising:

• receiving patient monitoring data from a patient monitoring device, wherein the patient monitoring device comprises at least one photodiode or at least one electrode coupled to an analog-to-digital converter (ADC) and at least one of a hardware filter or an amplifier to convert analog signals into digital signals and reduce motion and sensor noise representing patient monitoring parameters comprising electrocardiogram (ECG) data, a heart rate, a blood pressure, an oxygen saturation, a respiration rate, a temperature, or a combination thereof;
• storing a patient-specific data bank indexed by at least patient identity, acquisition time, and originating monitor identifier and configured to store historical physiologic data and selected patient monitoring data indicative of false alarms;
• filtering, based on the digital signals and a comparison from a machine learning module, a true alarm from a false alarm in real time, wherein the machine learning module comprises a contrastive neural network adapted to quasi-stationary physiological signals trained by a data sample collection, wherein the machine learning module is configured to (i) extract intermediate representations from the digital signals representing the patient monitoring data and corresponding historical physiological data, (ii) compute a contrastive loss between said intermediate representations, and (iii) compare the digital signals representing the patient monitoring data against the historical physiological data and the selected patient monitoring data indicative of the false alarms stored in the patient-specific data bank to distinguish the false alarm from the true alarm based on whether the contrastive loss is less than or greater than a device-side threshold;
• reducing an audible alarm volume upon determination that the contrastive loss indicates the false alarm;
• emitting a distinct illumination pattern corresponding to the false alarm condition;
• wherein the method avoids any alarms based on data from another patient by enforcing a match between the patient monitoring data and the historical physiological data and the selected patient monitoring data indicative of the false alarms indexed to the same patient identity and originating monitor identifier in the patient-specific data bank using the contrastive neural network, and wherein the method generates alarm conditions according to patient-specific patterns learned by the contrastive neural network rather than applying uniform thresholds across multiple patients.

A non-transitory data memory encoding one or more executable routines, which, when executed by a data processing device, cause the data processing device to perform acts comprising:

• receiving a patient monitoring data from a patient monitoring device, wherein the patient monitoring device comprises at least one photodiode or at least one electrode coupled to an analog-to-digital converter (ADC) and at least one of a hardware filter or an amplifier to convert analog signals into digital signals and reduce motion and sensor noise, wherein the patient monitoring data comprises patient monitoring parameters comprising electrocardiogram (ECG) data, a heart rate, a blood pressure, an oxygen saturation, a respiration rate, a temperature, or a combination thereof;
• storing a patient-specific data bank indexed by at least patient identity, acquisition time, and originating monitor identifier and configured to store historical physiologic data and selected patient monitoring data indicative of false alarms;
• filtering, based on the digital signals and a comparison from a machine learning module, a true alarm from a false alarm in real time, wherein the machine learning module comprises a contrastive neural network adapted to quasi-stationary physiological signals trained by a data sample collection, wherein the machine learning module is configured to (i) extract intermediate representations from the digital signals representing the patient monitoring data and corresponding historical physiological data, (ii) compute a contrastive loss between said intermediate representations, and (iii) compare the digital signals representing the patient monitoring data against the historical physiological data and the selected patient monitoring data indicative of the false alarms stored in the patient-specific data bank to distinguish the false alarm from the true alarm based on whether the contrastive loss is less than or greater than a device-side threshold;
• reducing an audible alarm volume upon determination that the contrastive loss indicates the false alarm;
• and
• emitting a distinct illumination pattern corresponding to the false alarm condition;
• wherein the method avoids any alarms based on data from another patient by enforcing a match between the patient monitoring data and the historical physiological data, and the selected patient monitoring data indicative of the false alarms indexed to the same patient identity and originating monitor identifier in the patient-specific data bank using the contrastive neural network, and wherein the method generates alarm conditions according to patient-specific patterns learned by the contrastive neural network rather than applying uniform thresholds across multiple patients.