AI-driven echocardiographic phenotyping in patients with heart failure – A help to identify amyloidosis and aortic stenosis in real word hospital cohorts

Background Heart failure (HF), valvular disease and cardiac amyloidosis are often underdiagnosed or detected late, particularly HF with preserved ejection fraction (HFpEF) and cardiac amyloidosis due to phenotypic overlap and operator-dependent echocardiography. Aortic stenosis in HF with reduced ejection fraction (HFrEF) patients is frequently missed because of low flow. Early diagnosis improves prognosis in both aortic stenosis and cardiac amyloidosis. Purpose To evaluate whether automated AI-based echocardiographic analysis improves detection of cardiac amyloidosis and aortic stenosis in real-world hospital cohorts. Method Retrospective 2148 randomly chosen patients who underwent echocardiography for various clinical indications in 2024 were included. An AI model (US2AI) certified for HF and amyloidosis was employed to identify patients with either; 1) cardiac amyloidosis or 2) HFrEF and aortic stenosis. Data was manually validated via electronic health records to assess diagnostic accuracy, management and outcomes. Results Among 2,148 echocardiographic reports, findings suggestive of heart failure were identified in 50.6% (n=1,087). Patterns with EF ≥50% predominated (n=848, 78.1%), including HFpEF and HFimpEF (Heart Failure with improved Ejection Fraction), HFrEF (n=129; 11.9%) and Heart Failure with mildly reduced Ejection Fraction (HFmrEF) (n=110; 10,1%). In the HFrEF subgroup (n=129; 11.9% of), the AI model detected 8 cases (6.2%) of moderate to severe aortic stenosis including 6 moderate and 2 severe; Of these, 5 participants underwent valve intervention, whereas 3 died shortly after diagnosis. Within the group with EF ≥50% (n=848), the AI model flagged 34 patients (3.2%) with features suggestive of cardiac amyloidosis. Only 2 participants (7.4%) had a prior amyloidosis diagnosis. Among the 32 subjects flagged for amyloidosis; 40.7% presented conduction abnormalities including atrial fibrillation, atrioventricular block, or bifascicular block and 4 had died. Conclusion Systematic AI-phenotyping of routine echocardiograms identifies and may help with underdiagnosed conditions such as; cardiac amyloidosis and aortic stenosis among individuals with HF. Integrating AI into clinical practice could fundamentally improve the patient-tailored treatment and avoid the irreversible clinical deterioration.