An Efficient Deep Feature Fusion Framework for Multi-Class Myocardial Infarction Detection Using ECG Signals

Abstract

Myocardial Infarct (MI) is a dangerous heart condition that needs to be diagnosed and treated right away. Automatic ECG-based diagnostic systems offer considerable potential; however, numerous current methodologies encounter challenges related to multi-class classification while computational efficiency. This research presents an effective deep feature fusion structure to multi-class myocardial infarction detection utilizing ECG signals. The framework combines advanced signal preprocessing, learning morphological features, and ensemble classification methods to make detection more accurate. Wavelet-based filtering as well as adaptive methods of normalization are used to remove noise from ECG signals at first. To find cardiac cycles, R-peak detection is used, and then beat segmentation is done. A stacked Auto-Encoder architecture-based deep feature extraction module learns compact connection representations that capture small pathological changes. These features learned are combined with statistical adjectives to make it easier to tell the difference. We look at a number of classifiers, such as Support Vector Machines and Gradient Boosting. Additionally, an extension method using Attention-Enhanced Features Reweighting is presented to continually prioritize clinically important features. The results of the experiment show that classification precision as well as durability have improved a lot. The suggested system provides a scalable, effective but and clinically pertinent solution for ECGbased myocardial infarction diagnosis.