Heart failure & cardiomyopathiesResearch article
Diagnosis of Cardiac Amyloidosis on Echocardiography Using Artificial Intelligence
- develop and validate an artificial-intelligence (AI)-based deep-learning model for video-based detection of cardiac amyloidosis (CA) on echocardiography
- external test data sets
Authors' conclusion
Both the multiparametric echocardiographic score computed from AI-derived measurements and the fully automated deep-learning model can accurately identify patients with CA in globally diverse cohorts, with the deep-learning model providing superior performance.
Study context
- Population
- Patients with cardiac amyloidosis (CA) and controls, including left ventricular hypertrophy controls, from the UK, Taiwan, US and JapanCA, 2756, controls, 3020 · Training data set (UK National Amyloidosis Center and Taiwan MacKay Memorial Hospital): CA 2,241 · Training data set: controls 2,130 · US Duke University Health System: CA 334 · US Duke University Health System: left ventricular hypertrophy controls 668 · Japan National Cerebral and Cardiovascular Center: CA 181 · Japan National Cerebral and Cardiovascular Center: left ventricular hypertrophy controls 222
- Participants
- 5,776 patients
- Comparison
- The deep-learning model compared with the AI-derived multiparametric echocardiographic score
Further findings
AUC
| Outcome | deep-learning model | AI-derived multiparametric echocardiographic score | P value |
|---|---|---|---|
| Diagnostic accuracy, deep-learning model versus AI-derived multiparametric echocardiographic score · superior diagnostic accuracy with the deep-learning model | 0.93 (95% CI 0.91–0.95) | 0.88 (95% CI 0.85–0.90) | P < 0.001 |
- Multiparametric echocardiographic score computed using AI-derived measurements: United States cohort accuracy 79.5% · sensitivity 75.4% · specificity 81.5%
- Multiparametric echocardiographic score computed using AI-derived measurements: Japan cohort accuracy 79.7% · sensitivity 81.6% · specificity 78.1%
- Deep-learning model, internal validation and internal test sets: internal validation accuracy 96.2% · sensitivity 96.8% · specificity 95.7%
- Deep-learning model, internal validation and internal test sets: internal test accuracy 95.8% · sensitivity 97.3% · specificity 94.3%
- Deep-learning model, external validation: United States accuracy 87.5% · sensitivity 86.6% · specificity 87.9%
- Deep-learning model, external validation: Japanese cohort accuracy 88.4% · sensitivity 92.3% · specificity 85.3%
Deep-learning model discrimination of CA from other hypertrophic phenocopies (subgroup analysis)
- CA versus hypertension AUC
- 0.92 [95% CI, 0.91-0.94]
- CA versus hypertrophic cardiomyopathy AUC
- 0.91 [95% CI, 0.87-0.94]
- CA versus aortic stenosis AUC
- 0.93 [95% CI, 0.90-0.95]
- CA versus chronic kidney disease AUC
- 0.93 [95% CI, 0.91-0.95]
Diagnostic accuracy, deep-learning model versus AI-derived multiparametric echocardiographic score
- The deep-learning model was able to classify a greater proportion of patients compared with the AI-derived multiparametric echocardiographic score
Original publication
Diagnosis of Cardiac Amyloidosis on Echocardiography Using Artificial Intelligence. Circ Cardiovasc Imaging. Published 12 February 2026. doi:10.1161/circimaging.125.018991
Summarised from the published abstract; every number is checked against it