Cardiovascular Disease Detection Based on Multi-Modal Data Fusion and Multi-Branch Residual Network

Cardiovascular diseases are commonly detected using bioelectrical signals such as electrocardiogram (ECG) and phonocardiogram (PCG), which reflect the state of the heart from different perspectives. However, previous studies on cardiovascular disease detection are mainly based on single-modal data, i.e. ECG or PCG alone. With the fast development of deep learning, researchers begin to pay attention to the detection of cardiovascular diseases using multi-modal data. In this study, we propose a multi-branch residual network that can automatically extract deep features from ECG and PCG signals. Different residual branches (SE-ResNet) can extract features at different scales. We use PCA to select the fused features and apply SVM classifier for classification. The experimental results demonstrate that the accuracy of our proposed method is 93.1% with an AUC value of 0.967, which outperforms methods using single-modal data as well as existing studies using multi-modal data. Our findings confirm that ECG and PCG signals are complementary in cardiovascular disease detection.