Self-Attention-Based Deep Convolutional Sparse Dense Autoencoder Model With Chebyshev-Based Osprey Algorithm for Cardiovascular Disease Detection

Abstract

Cardiovascular disease (CVD) refers to disorders affecting the heart and blood arteries. Automated screening methods can be used to identify CVDs, the leading cause of death worldwide. Electrocardiogram (ECG)-based techniques are widely utilized to detect CVDs since they are both noninvasive and efficient. This paper presents a deep convolutional neural network (CNN) for categorizing five CVDs using conventional 12-lead ECG data. The proposed approach comprises three steps: preprocessing, feature extraction, and classification. Initially, input signals are collected from a publicly available dataset; then, pre-processing is performed using a windowed infinite impulse response notch filter (W-I²RNF) to remove unwanted noise. The appropriate features are extracted from the preprocessed signals using the mel frequency cepstrum coefficient (MFCC) and modified discrete wavelet transform (Mod-DWT). CVDs are detected and classified based on the retrieved features using a new self-attention-based deep convolutional sparse dense autoencoder (SA_DC_SDAE) model. A deep CNN is combined with a sparse dense autoencoder (AE) technique to enable classification tasks. Also, the parameters of the proposed deep learning model are optimized using the Chebyshev-based Osprey Algorithm (C-OA). Therefore, the proposed model efficiently classifies CVDs with an accuracy range of 98.75%, sensitivity of 97.9%, precision of 95%, F1 score of 96%, and specificity of 99% for the PTB-XL dataset. The proposed model outperforms the state-of-the-art models in terms of performance.