London - ALsharqiya March 3: A machine learning algorithm has been developed that runs on a smartwatch, capable of detecting sudden loss of pulse with high accuracy. Add an ad The developed system aims to automatically monitor cardiac arrest cases, with the ability to make an emergency call when the case is detected, even if the user is unresponsive.
Out-of-hospital cardiac arrest (OHCA) is a major cause of sudden cardiac death, where survival chances depend on the speed of detection and medical intervention. Since 50-75% of these cases occur without witnesses, the patient is less likely to receive an immediate medical response.
To address this problem, the researchers sought to verify whether the smartwatch could independently detect loss of pulse, and contact emergency services while reducing false alarm rates.
Testing the algorithm In this regard, the researchers used photoplethysmography (PPG) data and motion measurements to train the algorithm, then tested it across 6 different groups, including controlled clinical environments and real-world conditions.
In the electrophysiology lab, 100 patients underwent defibrillator testing, allowing data on pulseless events to be recorded. Another 99 volunteers participated in the pulseless experiment via a tourniquet (a medical device used to compress a limb to prevent blood flow). A group of 948 users provided additional data without recording any pulseless events.
Test results 220 participants wore the smartwatch during their daily lives to assess the false alarm rate, while 135 people underwent tests in controlled environments, where the pulse was intentionally stopped via an arterial blockage to assess the sensitivity of the algorithm. 21 trained people also simulated cardiac arrest collapses outside the hospital to test the accuracy of the algorithm.
The results revealed no statistical difference between PPG signals resulting from ventricular fibrillation and pulseless events resulting from arterial blockages.
The algorithm’s sensitivity to cases where there was no pulse or movement was 72%, while the sensitivity to simulated collapse cases was 53%. The specificity rate reached 99.99% (the algorithm’s ability to avoid false alarms), with the watch rarely accidentally calling emergency services. The system was also able to detect loss of pulse within 57 seconds, with the user’s response checking mechanism activated for 20 seconds before making the emergency call.
