Efficient Siamese Network Based Approach for Multi-Class ECG Classification in Arrhythmia Detection

Accurately categorizing electrocardiogram (ECG) signals is fundamental for the early detection and intervention of cardiovascular diseases, notably arrhythmias. Traditional classification algorithms employ convolutional neural networks, in its essence, which itself have proved to be the state of art solution, but only from a single perspective. The stellar results provided by such solutions are still reproducible, provided large balanced datasets and huge computational resources are at hand. Failure in provision of the latter may lead to poor generalization of the model. These challenges called for a revision in the traditional methods, which motivated in the employment of Few-Shot learning based Siamese Convolutional Neural Networks (SCNNs) in this methodology. Unlike traditional classification models, the Siamese network focuses on identifying subtle differences between ECG patterns, enhancing its performance on imbalanced datasets and improving generalization across patient-specific variations. This paper proposes a novel supervised deep learning approach using Siamese Networks to classify ECG heartbeats sampled from the Massachusetts Institute of Technology and Beth Israel Hospital (MIT-BIH) Arrhythmia database. The proposed model presented an accuracy of up to 97.5%• These experimental results highlight the model's superiority in handling imbalanced datasets and its potential for practical application in real-time ECG monitoring systems.