Efficient Compressed Sensing for Real-Time Electrocardiogram Acquisition on Low-Power Implanted Medical Devices

Mitchell Terrell, Jon Weissman

Keywords: compressed sensing, electrocardiogram, tailored sensing, signal reconstruction, implantable medical devices

TL;DR: A novel compressed sensing method is proposed for real-time EKG acquisition on implanted medical devices which has high reconstruction accuracy with an extremely low computational burden on the device.

Abstract: Low cost wearable and implantable cardiac monitoring devices (WCM \& ICM) paired with increasingly accurate disease and arrhythmia detection algorithms have proven effective to help slow the impact of cardiovascular disease, the worlds leading cause of death in 2022. Improvements to server-side detection algorithms along with hardware limitations of these devices such as slow processor speeds and minimal battery life has led to a desire to offload data from the devices for later analysis. Paired with limited storage capacity, this has led to a push to decrease the necessary storage for electrocardiogram (EKG) signals without sacrificing disease detection accuracy and device longevity. A promising recent innovation in EKG compression has come from Compressed Sensing (CS), which exposes inherent sparseness in the signal to selectively sample for eventual server-side reconstruction. Many CS approaches have been implemented on WCM devices which demonstrate a high compression ratio (CR) and accurate signal reconstruction, but quickly become impractical for the stricter hardware constraints of ICM devices due to the increased computation on the device and slow reconstruction time. In this paper we propose a CS approach known as Tailored Sensing (TS) which combines on-device prior knowledge, a custom transform basis, and optimal sense location selection to achieve improved CR and reconstruction accuracy while eliminating on-device computational burden and slow reconstruction time. Our approach offers equivalent or better signal reconstruction at previously unfathomable CRs due to our novel signal segmentation scheme. Additionally, our approach boasts a zero overhead on-device sensing strategy and a 93\% reduction in signal reconstruction time.

Track: 4. AI-based clinical decision support systems

Registration Id: QWNBFJRM6SF

Submission Number: 260