A Framework for Temporal Learning Model for Early Myocardial Infarction Localization Using Multi-Lead ECG Signals

Abstract

Timely the localization of myocardial Infarction (MI) is very important for making clinical decisions and keeping patients alive. Traditional ECG-based diagnostic devices frequently prioritize static feature analysis, neglecting the temporal dependencies intrinsic to cardiac signals. This paper presents a lightweight a temporal learning framework intended for the early localization of myocardial infarction utilizing multi-lead ECG signals. Adaptive filtering methods are built into the framework to get rid of noise and distortions in the baseline. Temporal segmentation is used to keep the sequential dependencies in ECG cycles. We use a recurrent neural construction on the basis of gates recurrent units (GRU) to model how signals change over time. Learned temporal traits are integrated with particular domain morphological indicators to improve classification accuracy. The main contribution is the introduction of a method called Temporal Consistency Regularization (TCR). This extension makes sure that predictions are stable across close ECG segments, which makes them more general and robust. Tests show that the accuracy of localization has improved a lot and the variance in predictions has gone down. The suggested system is a simple, effective way to monitor the heart in real time using ECG data.