ECG Based Fusion Model Deep Learning and Fuzzy Logic for Precise Arrhythmia Identification and Risk Prediction in Drug Induced and Long Qt Syndrome

Abstract

This research presents a sophisticated and comprehensive framework for the automated detection and classification of cardiac arrhythmias, specifically focusing on both Cardiac Arrhythmia (CA) and Drug-Induced Arrhythmias (DA). Cardiac Arrhythmia is characterized by irregular heart rhythms, which may manifest as excessively slow or rapid beats, posing a significant risk of Sudden Cardiac Death (SCD). Given its potentially fatal implications, early detection is critical in mitigating mortality rates. While various methodologies have been developed for ECG beat classification, existing techniques often exhibit limitations in fiducial point detection, leading to misclassification. To address these shortcomings, this study introduces a novel Memory Layer with a Logistic BCM function-based Deep Learning Neural Network (ML-DLNN) model for CA detection. newlineInitially, the input ECG signals undergo refinement using the First Order Derivation-based Kaiser Derived Bessel (FOD-KDB) method to eliminate noise and correct the isoelectric line. Wave detection is performed utilizing the PTA technique, followed by decomposition via Deviation Measure-based Ensemble Empirical Mode Decomposition (DM-EEMD), ensuring precise identification of fiducial points. The fiducial point analysis facilitates the computation of relationships between various cardiac intervals, followed by feature extraction. The most relevant features are subsequently selected using the Frechet Distribution-based Lemurs Optimization Algorithm (FD-LOA), which are then fed into the ML-DLNN classifier. Experimental validation of the proposed framework demonstrates superior detection accuracy compared to conventional approaches. newlineFurthermore, this study investigates Drug-Induced Arrhythmias, abnormal heart rhythms triggered by specific medications that adversely affect cardiac functionality. Existing classification models predominantly rely on ECG signal analysis but fail to incorporate the consequences associated with DA newline