Recognition of Child Congenital Heart Disease using Electrocardiogram based on Residual of Residual Network
Yunmei Du, Shuai Huang, Canhui Huang, Allam Maalla, Huiying Liang
Abstract
Congenital heart disease seriously affects children's physical and mental health. Early screening is of great significance. Inexpensive, noninvasive and painless ECG combined with artificial intelligence technology can show new information about heart disease and help CHD screening. Using child ECG from 68969 patients at the GZFEZX, we trained a deep neural network model with two-level residual to identify CHD patients based on ECG. Experiments show that the two-level residual structure has better performance than the traditional ResNet. In the independent test set, the accuracy of the model is 92.30%, the sensitivity is 74.73%, and the specificity is 94.07%. The performance exceeds other individual CHD screening indicators, which shows that ECG is of great value for CHD identification and can be considered to be included in the screening process.