Detection of cardiovascular disease using deep learning-based attention-guided bi-LSTM with electrocardiogram signals

Abstract

Cardiovascular illnesses are one of the most dangerous conditions affecting the human heart today. To forecast and diagnose cardiac disease, medical experts and clinical data analysis face a serious challenge. In this work, a novel Deep learning-based Attention-Guided Bi-LSTM was proposed for the early detection of heart disease. The input signals are initially pre-processed using SWT Filtering, which is utilized to denoise the ECG signals. The input signal is initially split up into lower and higher frequency components by the stationary wavelet transform. The appropriate Gaussian filter is then applied separately to each of the deconstructed components. In the end, the components with low and high frequencies are reverse-transformed to produce the optimized signal. The attention-guided Bi-LSTM network is then built with an attention module to focus on the feature extraction stage to obtain enriched features without losing the relevant features before concatenation. The layer that has the Gelu activation function fully attached obtains the fused features to accurately categorize the normal and abnormal patients from the ECG signal. Early cardiovascular disease detection achieves an average classification accuracy of 99.34%. The proposed Attention-Guided Bi-LSTM improves overall accuracy by 0.94%,0.55%,0.26% and 0.34% over the SMOTE Technique, SMOTE-ENN, XGBoost, and DBSCAN, MOWPT, and CQP, respectively