Deep learning-based measurement of echocardiographic data and its application in the diagnosis of sudden cardiac death.

Abstract

This study aimed to evaluate the potential of deep learning applied to the measurement of echocardiographic data in patients with sudden cardiac death (SCD). 320 SCD patients who met the inclusion and exclusion criteria underwent clinical evaluation, including age, sex, BMI, hypertension, diabetes, cardiac function classification, and echocardiography. The diagnostic value of deep learning model was observed by dividing the patients into two groups: training group (n=160) and verification group (n=160), as well as two groups of healthy volunteers (n=200 for each group) during the same period. Logistic regression analysis showed that MLVWT, LVEDD, LVEF, LVOT-PG, LAD, E/e' were all risk factors for SCD. Subsequently, a deep learning-based model was trained using the collected images of the training group. The optimal model was selected based on the identification accuracy of the validation group and showed an accuracy of 91.8%, sensitivity of 80.00%, and specificity of 91.90% in the training group. The AUC value of the ROC curve of the model was 0.877 for the training group and 0.995 for the validation groups. This approach demonstrates high diagnostic value and accuracy in predicting SCD, which is clinically important for the early detection and diagnosis of SCD.