Voice-based monitoring and alerting for remote decompensated heart failure detection
A machine learning based patient voice monitoring and analysis system can reduce the need for patient hospitalization by early detection and treatment of health conditions such as acute decompensated heart failure.
1. A detecting and alerting system for detecting onset of decompensated heart failure in remotely-located human subjects and automatically generating an electronic alert based on a detected onset of decompensated heart failure in a remotely-located human subject, the detecting and alerting system comprising:
a data receiver that receives digitized voice samples of remotely-located human subjects;
a computing instance operatively connected to the data receiver, the computing instance configured to perform operations comprising:
(i) executing a first machine learning model trained on voice samples from a population to detect at least one feature in speech or voice characteristics that is a sign of onset of decompensated heart failure,
(ii) executing a second machine learning model that analyzes changes in acoustic features of voice samples received from the remotely-located human subject over time,
(iii) blending a first output from the first machine learning model based on voice samples received from the remotely-located human subject with a second output from the second machine learning model based on the voice samples received from the remotely-located human subject,
(iv) detecting onset of decompensated heart failure in the remotely-located human subject based on the blended first and second output, and
(v) generating a detection signal upon detecting onset of decompensated heart failure in the remotely-located human subject; and
an electronic alert system operatively connected to the computing instance, the electronic alert system configured to automatically generate an electronic alert in response to the computing instance generating the detection signal, the electronic alert suggesting that the remotely-located human subject exhibits signs of onset of decompensated heart failure.
2. The detecting and alerting system of claim 1 wherein the second machine learning model is trained on the remotely-located human subject.
3. The detecting and alerting system of claim 1 wherein the blending comprises majority voting, plurality voting, or computing a weighted averaging of results from the first machine learning model and the second machine learning model.
4. The detecting and alerting system of claim 1 wherein the data receiver receives digitized voice samples from telecommunication devices including smartphones of the human subjects.
5. A detecting and alerting method for detecting onset of decompensated heart failure in remotely-located human subjects and automatically generating electronic intervention alerts based on detected onset, the detecting and alerting method comprising:
receiving digitized voice samples of remotely-located human subjects;
using a computing instance, performing operations comprising:
(i) executing a first machine learning model trained on voice samples from a population to detect at least one feature in speech or voice characteristics that is a sign of onset of decompensated heart failure,
(ii) executing a second machine learning model that analyzes changes in acoustic features of received voice samples from a remotely-located human subject over time,
(iii) blending a first output from the first machine learning model based on voice samples received from the remotely-located human subject with a second output from the second machine learning model based on the received voice samples from the remotely-located human subject,
(iv) detecting onset of decompensated heart failure in the remotely-located human subject based on the blended first and second output, and
(v) generating a detection signal upon detecting onset of decompensated heart failure in the remotely-located human subject; and
automatically generating an electronic alert in response to the computing instance generating the detection signal, the electronic alert suggesting that the remotely-located human subject exhibits signs of onset of decompensated heart failure.
6. The detecting and alerting method of claim 5 wherein the second machine learning model is trained on the remotely-located human subject.
7. The detecting and alerting method of claim 5 wherein blending includes computing a weighted average of results of the first machine learning model and the second machine learning model.
8. The detecting and alerting method of claim 5 wherein receiving includes receiving digitized voice samples produced by a smartphone.
9. The detecting and alerting method of claim 5 further including the computing instance storing labelled received digitized voice samples detected as indicative of onset of decompensated heart failure for application to the first machine learning model or the second machine learning model.