Advancements in VLSI Based Neural Network Architectures for Abnormal Heartbeat Detection: A Comprehensive Review

Abnormal heart rhythms are an indication of cardiovascular diseases and account for most deaths worldwide. Detection of such abnormalities at an early stage can avoid severe complications, such as sudden cardiac death. Electrocardiogram monitoring has long been the standard practice in clinics to diagnose arrhythmia because that method remains confined to clinics and can offer only periodic assessment capabilities. Advances in VLSI technology have led to the design of very compact, low power devices for real-time ECG analysis to support continuous cardiac monitoring in wearable health devices. At the same time, deep learning architectures, such as CNNs and DSNNs, have been shown to attain great promise in the automatic detection of arrhythmic patterns from ECG data. This work deals with an extensive literature review for VLSI-based neural network architectures for the detection of heartbeat anomalies using ECG, specifically emphasizing CNN and DSNN, where their structural designs, power efficiencies, classification accuracy, and integrations into handheld devices have been compared. The CNN-based designs tend to provide complicated feature extraction using deep learning methods, while DSNNs incorporate a novel data-shifting technique that maximizes the model's accuracy without any additional increase in circuit complexity. This paper evaluates and compares critical performance metrics such asaccuracy, power consumption, and chip area. As we discuss further on these points, this work covers the ways in which VLSI-enabled neural networks are applied to wearable ECG monitoring systems and designs future guidelines that could enhance real-time healthcare solutions.