Patricia A. Pellikka, MD, FACC, FAHA, FASE, FESC (and team)
Mayo Clinic, MN, United States of America
Read the slides from the full presentation at ESC Congress 2024
Background
Cardiac amyloidosis is associated with high morbidity and mortality, with transthyretin (ATTR) amyloidosis being more common than previously thought. Its presentation is variable, and early diagnosis is critical, as transthyretin stabilizers are most effective in the early stages. Echocardiograms of these patients often present complexities, leading to missed diagnoses.
Methods
The study aimed to develop an automated echocardiographic screening tool for cardiac amyloidosis using a single videoclip of the apical four-chamber view. The tool was trained using a 3D convolutional neural network on patients diagnosed with AL cardiac amyloidosis, ATTR cardiac amyloidosis, hypertrophic cardiomyopathy (HCM), aortic valve stenosis, and other similar conditions. The external validation cohort included a diverse, global cohort of 597 patients with cardiac amyloidosis and 2122 controls.
Results
In the tuning dataset (7666 patients), the model achieved an area under the curve (AUC) of 0.928, with 81.4% sensitivity and 92.8% specificity. In external validation on 2719 patients, the model performed similarly well, with an AUC of 0.93, 85% sensitivity, and 93% specificity. The model's performance was consistent across various subgroups, such as AL, ATTR wild type, and ATTR variant. In a subgroup of patients matched for age, sex, and wall thickness, the AUC ranged from 0.86 to 0.92.
Conclusion
The AI screening model demonstrated strong performance in detecting cardiac amyloidosis across diverse patient populations, outperforming the traditional Transthyretin Cardiac Amyloid Score in patients over 60 years of age with heart failure and increased LV wall thickness. This model has the potential to improve early detection of cardiac amyloidosis, facilitating timely access to treatment.
