International Validation of Echocardiographic Artificial Intelligence Amyloid Detection Algorithm.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Cardiac amyloidosis (CA) diagnosis is often delayed due to its similarity to other conditions causing left ventricular wall thickening.
• Conventional echocardiography has limitations in specificity for CA detection.

Purpose of the Study:

• To evaluate the performance of a computer vision algorithm, EchoNet-left ventricular hypertrophy (LVH), for identifying cardiac amyloidosis (CA).
• To assess the algorithm's diagnostic accuracy across multiple international sites in a case-control study.

Main Methods:

• A retrospective case-control study involving 574 CA patients and 979 controls.
• Utilized EchoNet-LVH, a deep learning algorithm analyzing echocardiogram videos (parasternal long axis and apical-4-chamber views).
• Evaluated diagnostic performance using area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and positive predictive value.

Main Results:

• EchoNet-LVH achieved an AUC of 0.896.
• At a threshold optimized for specificity, the algorithm demonstrated a sensitivity of 0.644 and a high specificity of 0.988.
• Positive predictive value was 0.968, and negative predictive value was 0.828, with consistent performance across various patient demographics and study sites.

Conclusions:

• EchoNet-LVH shows potential to assist in earlier and more accurate diagnosis of cardiac amyloidosis.
• The algorithm's high specificity is crucial for maximizing the positive predictive value of confirmatory tests in rare diseases like CA.