Spiking Neural Networks for ECG Classification and Anomaly Detection

Spiking Neural Networks (SNNs) have emerged as biologically inspired models that enable event driven, energy efficient computation, making them particularly suitable for biomedical signal processing in resource-constrained environments. Conventional machine learning and deep learning approaches, while effective, often struggle to capture the temporal dynamics of electrocardiogram (ECG) signals and typically demand high computational resources, which limits their applicability in wearable and real-time healthcare systems. To address these challenges, this study introduces a reproducible end to end framework for ECG classification and anomaly detection using a lightweight SNN architecture based on Leaky Integrate and Fire neurons. The framework integrates pre-processing, spike encoding, supervised training, and evaluation, ensuring methodological clarity and replicability. Validated across three benchmark datasets MIT-BIH Arrhythmia, PTB Diagnostic ECG, and ECG Heartbeat Categorization the proposed model achieved accuracies of 90.35%, 93.25%, and 97.8 % respectively. Despite its simplicity, the framework consistently outperformed or matched more complex deep learning and traditional machine learning methods, while requiring far fewer computational resources. These findings highlight the capacity of SNNs to effectively capture temporal and morphological features of ECG signals, enabling reliable detection of arrhythmias and myocardial infarctions. By combining clinical relevance with algorithmic transparency, the study addresses reproducibility gaps in neuromorphic research and demonstrates the feasibility of deploying lightweight SNNs in real-time, low-power cardiac monitoring applications.