Enhanced Cardiac Arrhythmia Detection Utilizing Deep Learning Architectures and Multi-Scale ECG Analysis

Abstract

Cardiovascular disease is the leading cause of death globally, with arrhythmia being a particularly lethal condition. Efficient and accurate identification of arrhythmia through the analysis of ECG data is crucial for effective treatment. Arrhythmias must be assessed when examining ECGs. This study presents a novel approach to automatically diagnose arrhythmia and congestive heart failure from sinus rhythm. The proposed method involves utilizing a multi-scale filter bank with scalograms, which makes use of preprocessed ECG data and non-weighted, pre-trained convolutional neural networks. Temporal frequency textures provide two-dimensional representations of fundamental characteristics from single-lead ECG recordings. Subsequently, deep learning neural networks that are specifically designed for arrhythmia classification are used to label and classify collections of feature data. This study investigates the efficacy of deep learning models in classifying cardiac arrhythmias from ECG data. The study looks at how well different convolutional neural network architectures work by using a multi-scale filter bank and a scalogram-based representation. Pre-trained networks yielded classification models that were both 100% accurate and more effective than raw networks in terms of generalization. A comparison of models that have been trained and models that have not been trained shows that pre-trained networks, especially Vgg16, perform better in many ways, such as accuracy and precision. This suggests the potential for significant improvements in automated ECG-based diagnostics, paving the way for advanced, personalized healthcare solutions.