Automated Echocardiographic Detection of Congenital Heart Disease Using Artificial Intelligence
Authors' conclusion
EchoFocus-CHD shows promise for automated CHD detection to advance equitable global cardiovascular care
Study context
- Population
- Patients undergoing echo for congenital heart disease (CHD) detection (internal median age 7.1 years; referral median age 2.5 years)counts are echos
- Participants
- 54,727 vs 3,356
- Comparison
- Internal vs Referral
- Internaln = 54,727
- Referraln = 3,356
- Outcome
- Detection of composite critical CHD by EchoFocus-CHD
Methodological detailsAI model development, internal training (80%) and testing (20%), external referral cohort evaluation
- AI model development
- Internal training (80%) and testing (20%)
- External referral cohort evaluation
Performance
As reported in the abstract
Composite critical CHD detection
| Test or model | AUROC | Positive likelihood ratio | Negative likelihood ratio |
|---|---|---|---|
| Internal | 0.94 | 7.50 | 0.14 |
| Referral | 0.77 | Not reported in the abstract | Not reported in the abstract |
AUROC
Internal0.94
Referral0.77
Positive likelihood ratio
Internal7.50
ReferralNot reported in the abstract
Negative likelihood ratio
Internal0.14
ReferralNot reported in the abstract
Further findings
κ
| Outcome | Internal | Referral |
|---|---|---|
| Expert agreement | 0.82 | 0.72 |
- Individual critical lesions (internal)
- 0.83-1.00 · AUROC
- Composite noncritical CHD (internal)
- 0.90 · AUROC · Positive likelihood ratio 5.00 · Negative likelihood ratio 0.23
- Individual noncritical lesions (internal)
- 0.70-0.96 · AUROC
International critical CHD detection after retraining on all US patients
- AUROC
- 0.87
- Domain shift attenuated; calibration improved
Original publication
Automated Echocardiographic Detection of Congenital Heart Disease Using Artificial Intelligence. Circulation. Published 28 March 2026. doi:10.
Summarised from the published abstract; every number is checked against it