Establishment and efficacy evaluation of deep learning model for cardiac conduction system / 中国法医学杂志

ABSTRACT

Objective To investigate the recognition efficiency of AI model based on deep learning for cardiac conduction system(CCS).Methods HE staining sections of cardiac muscle and CCS of 17 cases of non-sudden death were selected,and the gold standard was unanimous recognition by 2 forensic pathologists with more than 20 years of CCS diagnosis experience.Inception V3 algorithm was used to establish AI model and complete CCS identification training and testing.Confusion matrix,accuracy,precision,recall,F1 score,ROC curve and AUC value were used to evaluate the effectiveness of AI model,and accuracy,sensitivity and specificity were used to evaluate the efficiency of manual independent and AI-assisted manual recognition for CCS.Results The accuracy of AI model was 87.3%,the precision was 91.9%,the recall was 81.9%,the F1 score was 86.6%,and the AUC value was 95.3%.The accuracy of AI model was higher than that of senior forensic pathologists.There was no statistical significance in the accuracy of AI-assisted senior forensic pathologists in identifying CCS compared with manual independent detection(P>0.05),while the accuracy of AI-assisted intermediate and junior forensic pathologists in identifying CCS was increased by 8%and 14.33%,respectively,with statistical significance(P<0.05).The accuracy rate of AI-assisted junior forensic pathologists to identify CCS was higher than that of intermediate forensic pathologists in self-diagnosis.Conclusion The AI model could be used for the automatic recognition of CCS,and could improve the diagnostic efficiency of CCS and narrow the gap between the forensic pathologists with low experience and that with high experience.