Advanced ensemble learning-based CNN-BiLSTM network for cardiovascular disease classification using ECG and PCG signal

Cardiovascular disease (CVD) is a well-known leading cause of death worldwide. This highlights the need for an effective and efficient diagnostic-therapeutic path for the diagnosis and risk stratification of coronary artery disease (CAD) patients. However, it is inaccurate to investigate CAD only based on either electrocardiogram (ECG) or phonocardiogram (PCG) recordings. Several studies have attempted to use a combination of both signals in the early prediction and diagnosis of CAD. Considering the strong capability of deep learning models in feature extraction this research explores the efficiency of a hybrid CNN-BiLSTM ensemble approach that combines ECG and PCG signals to determine cardiac health status. Inspired by the significant performance of ensemble learning techniques in combining multiple base models to enhance overall prediction accuracy, a hybrid network architecture is suggested. The proposed CNN-BiLSTM model is considered a baseline for ECG and PCG signal prediction. Then, a bilinear layer combines both predictions of individual models to obtain a final accurate and robust prediction. It applies a bilinear transformation to incoming outputs from two base models to make the final output. The proposed architecture shows considerable improvement in prediction accuracy compared to using both ECG and PCG signals separately. Employing the well-known PhysioNet/Computing in Cardiology (CinC) Challenge 2016 Database, the proposed method has achieved 97% diagnosis accuracy, which how improvement over comparable methods and various other existing techniques.