Congenital heart disease detection by pediatric electrocardiogram based deep learning integrated with human concepts
Problem
Detecting congenital heart disease in children early using low-cost, non-invasive pediatric ECG instead of more invasive or expensive methods.
Why it matters
Early detection of congenital heart disease is critical to improved treatment outcomes for child patients.
What the AI did
CHDdECG, a deep learning model, automatically extracts ECG and wavelet features, integrates human-concept features, and detects CHD from pediatric ECGs.
Results
ROC-AUC 0.915, specificity 0.881 on 12,000-case real-world test set; external test sets ROC-AUC 0.917 and 0.907, specificities 0.937 and 0.907; surpassed cardiologists.
Impact
External validation on two independent datasets supports robust CHD detection from routine ECG.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2024 |
| Work type | article |
| Authors | Jintai Chen, Shuai Huang, Ying Zhang, Qing Hua Chang, Yixiao Zhang, Dantong Li, Jia Qiu, Lianting Hu et al. |
| Citations | 76 (OpenAlex, at import time) |
| References | 68 |
| Model or method | CHDdECG (deep learning) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-024-44930-y |
| Open access | gold |
| Keywords | deep learning, ECG, congenital heart disease, pediatric, screening, ROC-AUC |
Importance rating
significant — thousands to millions affected
Abstract states early CHD detection is critical to child treatment outcomes; low-cost non-invasive ECG screening, validated on 65,869+ cases with external sets, could benefit pediatric populations broadly.