Prediction of cardiovascular diseases by integrating multi-modal features with machine learning methods

Electrocardiogram (ECG) and phonocardiogram (PCG) play important roles in early prevention and diagnosis of cardiovascular diseases (CVDs). As the development of machine learning techniques, detection of CVDs by them from ECG and PCG has attracted much attention. However, current available methods are mostly based on single source data. It is desirable to develop efficient multi-modal machine learning methods to predict and diagnose CVDs. In this study, we propose a novel multi-modal method for predicting CVDs based both on ECG and PCG features. By building up conventional neural networks, we extract ECG and PCG deep-coding features respectively. The genetic algorithm is used to screen the combined features and obtain the best feature subset. Then we employ a support vector machine to implement classifications. Experimental results demonstrate the performance of our method is superior to those of single modal methods and alternatives. Our method reaches an AUC value of 0.936 when we use multi-modal features of ECG and PCG.