Classification of Heart Sounds Based on Topological Data Analysis Method

Abstract

Topological data analysis (TDA) method could catch the rich geometric and topologic information of big data and find subtle differences between different signals. TDA method opens up a new way for biomedical data analysis. In this study, we applied TDA method for heart sound signals (PCG) classification. First, the sliding window method was used to build a point cloud. Then, the persistent barcode is extracted from the point cloud by using the topology technology Vietoris-Rips (VR) filtration. At last, GoogLeNet transfer learning model was applied for classifing. The proposed the model did work well on the 2016 PhysioNet/CinC challenge dataset, Se=99.30%, +P=99.57%, F1=99.44%, mAcc=99.47%. The results showed that TDA can be used for the analysis of physiological signals in large quantities. The proposed method in this study has opened a new space for the application of TDA methods in physiological signal analysis.