Table 1 Model performances on the internal test set from Center-A, an external test set from Center-B, and another external test set from Center-C

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

View full size image

  1. The reported performances for detecting CHDs or non-CHD were obtained with the CHDdECG model trained from scratch, while the performances for various CHD subtypes were obtained separately after the CHDdECG model was specifically fine-tuned. Notably, some subtypes were not included since their amounts were <0.5% of all the CHD cases or <10. The probabilistic predictions are represented by box-plots for better viewing of the distributions. Box-plot elements are defined as center line, median; box limits, upper and lower quantiles; whiskers, 1.5 × interquartile range; n = 2038 CHD cases, 300 CHD cases, and 2521 CHD cases from Center-A, Center-B, and Certer-C, respectively. Source data are provided as a Source Data file.
Back to article page