Heart murmur grade recognition based on omni-dimensional dynamic convolution and feature fusion
SHOU Du
HUANG Zhaohan
LI Huhao
LI Li
ZHAO Qijun
PAN Fan
Abstract：To address the issue of subjectivity and susceptibility to environmental interference in murmur grading diagnosis through manual auscultation,we constructed a heart murmur grade recognition model based on omni-dimensional dynamic convolution(ODConv)and feature fusion.Firstly,the ODConv module was adopted to capture rich context information.Secondly,time-domain and time-frequency domain features were integrated to improve the recognition performance of murmur levels.Finally,the experiments on the amplitude attenuation and enhancement of the first heart sound(S1)and the second heart sound(S2)were designed to verify the influence of the relative loudness of S1,S2 and the murmur on the murmur level recognition of the model.The verification results on the CirCor DigiScope dataset showed that this model not only referred to the relative loudness information of the murmur with S1 and S2 in the classification of murmur levels,but also achieved recall rate of 93.39%,53.70%and 73.46%in the Absent,Soft and Loud cate-gories,as well as F1 score of 94.07%,51.28%and 73.34%,respectively.The proposed model can accurately identify heart murmur severity,and provide crucial technical support for the automated recognition of heart murmur levels.
Keywords：Deep learningFeature fusionHeart murmur gradingLogarithmic gammatone spectrums
Publication Date：2025-12-30
Online Publishing Date：2026-09-11(First online date of this platform, not the publication date of the document)
Pages：8( 387-394 )