Abstract

We present a framework and a novel dataset for detecting loss of pulse cases using a smartwatch. Our methodology for gathering data involves inducing controlled pulse loss in the forearm within a clinical setting under medical supervision. To the best of our knowledge, this is the first multimodal dataset comprising photoplethysmogram (PPG) signals, inertial measurements (IMUs) that can be obtained from a commercially available smartwatch, and continuous recordings as ground truth, collected from 20 individuals. This dataset serves as a valuable resource for the research community to advance and verify techniques for detecting pulse loss and differentiate between the non-usage of smartwatches to prevent false alarms. Our presented framework consists of extracting several features from signals and employing machine learning models designed to differentiate between the emergency loss of pulse cases with the non-usage of smartwatches. The presented framework achieves a detection accuracy of up to 97.1%, with a false detection rate of 1.0 in an hour when evaluated using leave-one-subject-out cross-validation.