Rare Disease AdvisorAI Model Accurately Detects ATTR-CM From ECG Images

Investigators presented data at the American College of Cardiology 2026 Scientific Session showing that an artificial intelligence model trained on electrocardiogram images could identify transthyretin‑mediated amyloid cardiomyopathy (ATTR‑CM). The approach aims to improve detection of a condition that is frequently underdiagnosed in routine cardiology practice.

This work describes a diagnostic screening model, not a definitive diagnostic test. Findings are based on retrospective and prospective cohort validation and require confirmatory testing for clinical diagnosis.

Clinical Considerations

• The AI model, ECGi‑ATTR, was trained using ECG images from over 11,000 patients, including a subset with confirmed ATTR‑CM.
• In a validation cohort, the model demonstrated strong performance, with an AUROC approaching 0.9, and maintained reasonable sensitivity and specificity in patients with conditions that mimic ATTR‑CM.
• Performance remained consistent in external validation using the SCAN‑MP cohort, which includes Black and Hispanic patients with heart failure, supporting generalizability across diverse populations.
• The model relies solely on ECG images, enabling potential scalability without advanced imaging or biomarker testing.

Clinical Practice Impact

• Earlier suspicion: ATTR‑CM often remains unrecognized until late disease; AI‑assisted ECG screening may help flag patients earlier in the diagnostic pathway.
• Workflow integration: An ECG‑based tool could prioritize patients for confirmatory imaging or nuclear scanning, particularly in resource‑limited or underserved settings.
• Equity considerations: Validation in minority populations suggests potential value in reducing disparities in ATTR‑CM detection.
• Limitations: The model does not replace diagnostic confirmation and requires further evaluation before routine clinical deployment.

More in Artificial Intelligence (AI)