Hybrid ECG-Based System for Accurate and Lightweight Myocardial Infarction Detection and Localization

Abstract

Myocardial infarction (MI) is a heart disease where blood flow to part of the heart muscle is blocked, causing tissue death due to lack of oxygen. These situations lead to the sometimes-human death. If not early detected can lead to the serious situations. MI detection and localization can be useful for early prevention of its aggravation and related cardiac health complications. In Existing works, they have developed a lightweight & efficient ML- based system which automatically detects various types of myocardial Infarction using a single-lead ECG features. To improve the feature quality, an autoencoder-decoder is introduced to learn high level representations which are helpful for selecting highly correlated features. Later a combination of  two different algorithm, Random Forest & XGBoost is introduced to improving accuracy. Existing system achieves 96.75% accuracy. To enhance the performance futher, the proposed system introduces  a model that mainly focuses on generating highly optimized and over fitted representations of the ECG signals. This model includes a overfit Deep Neural Network (DNN) Architecture that is capable of learning complex and non-linear relationships within the dataset. Here this overfit DNN is intentionally trained in an overfitting mode to get even the most subtle variations in the myocardial infarction PTB-XL dataset, this overfitting DNN helps the model to achieve perfect metric scores within that environment. This proposed model achieves the accuracy of 99.90% and other metrices on test data.