Automatic Heart Sound Classification Using One Dimension Deep Neural Network.

Cardiovascular disease (CVD) is one of the life-threatening diseases. Many researchers handcrafted features of heart sound to analyze heart sound signals for CVD automatically and achieved great success. But the handcrafted features of heart sound might not fully represent the raw data and it might be useless and redundant. Then the computational resources might be wasted. In this paper, the one dimension deep neural network (1-D DNN) with low parameters is proposed to detect abnormal of Cardiovascular disease. The raw heart sound fragments segmented by sliding window of 3s are fed into the network to extract discriminative features and are classified to normal or abnormal. The 2016 PhysioNet challenge database is used for training and validating the proposed network. Proposed network only has 0.08 Mb parameters and achieves 94.6% classification accuracy. Compared to the related works on heart sound analysis for Cardiovascular disease detection. The proposed 1-D DNN provided comparable performance in heart sound classification without handcrafted feature and precise segmentation.