Fig. 2: Visualization of CHDdECG-activated segments for CHD subtypes using the Grad-GAM approach. | Nature Communications

Fig. 2: Visualization of CHDdECG-activated segments for CHD subtypes using the Grad-GAM approach.

From: Congenital heart disease detection by pediatric electrocardiogram based deep learning integrated with human concepts

Fig. 2

a–d Illustrated the classical ECG manifestations of subtype ASD, VSD, d-TGA, and TOF, respectively. e illustrated the ECG manifestations of subtype VSD but misidentified as non-CHD by CHDdECG. f illustrated a segment of non-CHD ECG manifestations but misidentified to be CHD by CHDdECG. The leads of ECG views are marked on the top right. The salient ECG segments were marked by blue (the darker blue indicated the more important segments). We further circled the typical manifestations in orange that were considered to be associated with subtypes of adult ECG by the cardiologists following previous research results25. The horizontal axis represents time (4 × 10−2 s), and the y-axis shows the amplitude of the electrocardiogram (μV). Source data are provided as a Source Data file.

Back to article page