Edge-Optimized Weightless Neural Network for Low-Power Wearable Arrhythmia Detection

Cardiovascular disease (CVD) remains the leading cause of death worldwide. The standard diagnostic approach for CVD relies on electrocardiogram (ECG) analysis [1], which records the heart's electrical activity over time. However, the manual interpretation of the rapidly growing volume of ECG data demands considerable medical expertise and resources. This challenge has driven the development of automated ECG analysis techniques utilizing machine learning. In this paper, we present an innovative approach to integrating machine learning into a wearable medical patch. Leveraging the efficiency and low hardware complexity of Weightless Neural Networks (WNNs) combined with 45 nm CMOS Integrated Circuit technology, we introduce a specialized WNN model, “arrDWNN,” for arrhythmia detection from ECG signals. The arrDWNN architecture represents a significant advancement in wearable cardiac monitoring, enabling the development of next-generation low-cost ECG patches [2], for real-time arrhythmia detection. This edgecomputing solution integrates sensor functionality directly within the patch, facilitating real-time analysis of ECG signals for the immediate identification of potentially life-threatening arrhythmias. The proposed approach offers multiple advantages over traditional methods, including lower power consumption, continuous monitoring capabilities, and enhanced accessibility. By providing a rapid, cost-effective tool for detecting and managing cardiovascular events, the arrDWNN-based patch has the potential to revolutionize clinical practice, particularly in under served and low-income communities with limited access to specialized cardiac care worldwide. The proposed arrDWNN model achieves an impressive average accuracy of 89 % on the MIT-BIH [3] arrhythmia database. The hardware implementation [4], designed with 45 nm CMOS technology, features a compact circuit with 11,443 NAND2-equivalent gates and occupies an area of only 12.4 <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mm}^{2}$</tex> while operating at under 4.24 mW of power. This research highlights the potential of efficient, low-power machine learning solutions in wearable medical devices, enabling real-time cardiac monitoring to support preventive care and early intervention. In turn, this can significantly enhance patient outcomes while helping to alleviate the globalburden of cardiovascular disease.