Abstract
BACKGROUND: Diagnosis of cardiac amyloidosis (CA) is often missed or delayed due to confusion with other causes of increased left ventricular wall thickness. Conventional transthoracic echocardiographic measurements like global longitudinal strain have shown promise in distinguishing CA, but with limited specificity. OBJECTIVES: We conducted a multisite retrospective case-control study to investigate the performance of a computer vision algorithm for CA identification across multiple international sites. METHODS: EchoNet-left ventricular hypertrophy (LVH) is a computer vision deep learning algorithm for the detection of CA based on parasternal long axis and apical-4-chamber view videos. We evaluated EchoNet-LVH's ability to distinguish between the echocardiogram studies of 574 CA patients and 979 controls. We reported discrimination performance with an area under the receiver operating characteristic curve and associated sensitivity, specificity, and positive predictive value at the prespecified threshold. RESULTS: EchoNet-LVH had an area under the receiver operating characteristic curve of 0.896 (95% CI: 0.875-0.916). At the prespecified model threshold optimizing for specificity, EchoNet-LVH had a sensitivity of 0.644 (95% CI: 0.601-0.685), specificity of 0.988 (95% CI: 0.978-0.994), positive predictive value of 0.968 (95% CI: 0.944-0.984), and negative predictive value of 0.828 (95% CI: 0.804-0.850). There was no evidence of heterogeneity in performance by site, race, sex, age, body mass index, CA subtype, or ultrasound manufacturer. CONCLUSIONS: EchoNet-LVH can assist with earlier and accurate diagnosis of CA. EchoNet-LVH achieved development goals to be highly specific to maximize positive predictive value of downstream confirmatory testing since CA is a rare disease.