Deep Learning-Based Electrocardiogram Analysis Predicts Biventricular Dysfunction and Dilation in Congenital Heart Disease
Authors' conclusion
AI-ECG shows promise to predict biventricular dysfunction/dilation, which may help inform CMR timing in CHD.
Study context
- Population
- Patients with and without congenital heart disease (CHD) with paired ECG-CMRs (≤30 days apart)
- Participants
- 4,941 vs 746
- Comparison
- Internal testing vs External validation
- Internal testingn = 4,941
- External validationn = 746
- Outcome
- AI-ECG detection of cardiovascular magnetic resonance (CMR)-defined LV dysfunction (EF ≤40%), RV dysfunction (EF ≤35%), LV and RV dilation (end-diastolic volume z-score ≥4), by area under receiver-operating curve (AUROC)
Methodological detailsmodel development and external validation
- Model development and external validation
Performance
As reported in the abstract
AUROC
| Test or model | Left ventricular (LV) dysfunction AUC | LV dilation AUC | RV dysfunction AUC | RV dilation AUC |
|---|---|---|---|---|
| Internal testing | 0.87 | 0.86 | 0.88 | 0.81 |
| External validation | 0.89 | 0.83 | 0.82 | 0.80 |
Left ventricular (LV) dysfunction AUC
Internal testing0.87
External validation0.89
LV dilation AUC
Internal testing0.86
External validation0.83
RV dysfunction AUC
Internal testing0.88
External validation0.82
RV dilation AUC
Internal testing0.81
External validation0.80
Further findings
Performance in functionally single ventricle (subgroup)
- Model performance lowest in functionally single ventricle patients
Survival in tetralogy of Fallot predicted high risk of ventricular dysfunction
- Lower survival in high-risk-predicted tetralogy of Fallot patients
- P < 0.001
Saliency mapping
- Lateral precordial leads influence all outcome predictions; QRS widening and T-wave inversions high-risk features for RV dysfunction/dilation
Original publication
Deep Learning-Based Electrocardiogram Analysis Predicts Biventricular Dysfunction and Dilation in Congenital Heart Disease. J Am Coll Cardiol. Published 27 August 2024. doi:10.
Summarised from the published abstract; every number is checked against it