【摘要】 心音信号的研究有助于先天性心脏病的早期辅助诊断。提出一种对先天性心脏病心音分类的新方法：对每例心音截取2秒作为样本；用经验模态分解、多正弦窗、幂函数压缩法对MFCC与IMFCC进行改进，并用改进后的MFCC与IMFCC分别提取心音样本相应频率系数，并计算各自的一阶差分作为融合特征。分类模型选用两层CNN网络。对5000例样本训练测试的二分类准确率为0.921,灵敏度和特异度分别为0.898、0.944;F1和AUC分别达到了0.919与0.958。上述法有望用于先心病机器辅助诊断。更多还原

【Abstract】 The study on heart sound signals is helpful for the early diagnosis of congenital heart disease. A novel method for heart sound classification of congenital heart disease was proposed in this paper. 2 seconds were taked from each heart sound as a sample; MFCC and IMFCC were improved by using empirical mode decomposition, multiple sine windows, and power function compression methods. The coefficients of sample were extracted by using improved MFCC and IMFCC methods. Their first-order differences were calculated respectively. All above were mixed as fusion features. A two-layer CNN network was used as classifier. 5000 samples were used in this study. The results show that an accuracy rate of 0.921,a sensitivity of 0.898,and a specificity of 0.944 are achieved using novel signal processing; F1 and AUC are of 0.919 and 0.958 respectively. The above method is expected to be used for machine assisted diagnosis of congenital heart disease.更多还原