Advanced Deep Learning Approaches for Automated Diagnosis of Cardiac Arrhythmia in Multi-lead ECG Signals

Abstract

Atrial fibrillation (AF) is a widespread and life-threatening heart arrhythmia disorder that affects over a million persons worldwide, predominantly older adults. The exact detection of AF is important, but the complication of ECG signals leads to conventional methods based on manual interpretation. To address this challenge, in this paper a new deep learning method has been proposed that used an Bidirectional Long Short-Term Memory (BLSTM) block with the deep Convolutional Neural Network (CNN). The processed data is analyzed through a four-layer deep learning model, consisting an BLSTM block of two and two layers that are fully connected. In this project, the first step is an analysis process involves wavelet transform for ECG signal preprocessing, that removed the noise and enhanced the signal clarity. Peak extraction and segmentation preformed. Two datasets signal, one for RR interval (dataset A) and one for sequence of heartbeats in terms of waves P-QRS-T, then after the fully connected layer it detects if the signal is AF or not an AF. The model performs remarkably, DenseNet-169 + BLSTM remains the best-performing model, achieving the highest accuracy (98.75%) while maintaining a good balance between precision, sensitivity, and specificity. To the best of understanding, when compared to many other methods, the suggested technique produces great result states of filters and algorithm, this provides an advance technique for the detection of AF. These values confirm the dependability and high efficiency of proposed AF detection about current AF detection methodologies.