AI-assisted echocardiography shows moderate agreement with cardiologists in ATTR-CM monitoring

Background: Detection of disease progression is key to personalize treatment strategies in transthyretin cardiomyopathy (ATTR-CM), particularly with emerging therapies. Echocardiography can detect subtle longitudinal changes but is limited by operator dependence. This study evaluates agreement and reproducibility of fully automated, AI-assisted echocardiographic measurements under real-world conditions. Methods: This retrospective study included 62 patients with ATTR-CM undergoing 178 serial annual echocardiograms assessed by a reference cardiologist, a second cardiologist, a novice reader, and a fully automated AI algorithm (Us2.ai). Interrater agreement was assessed using Bland-Altman analysis and intraclass correlation coefficients (ICCs). Intrarater variability for human readers was derived from repeated blinded measurements, with limits of agreement (LoA = mean difference +/- 1.96 x SD) defining the smallest detectable change. AI repeatability was assessed using within-study pairwise differences. Results: AI showed moderate agreement with the reference cardiologist for IVSd and LVEDV (ICC 0.65 and 0.51), with biases of -1.9 mm and -39 mL, respectively. Interrater agreement between cardiologists was good (ICC 0.79 and 0.84) with minimal bias (-0.2 mm and +3 mL). Intrarater variability was moderate to excellent for both cardiologists (LoA 3.0 mm and 43 mL for the reference cardiologist; 2.7 mm and 31 mL for the second cardiologist). AI demonstrated comparable repeatability (LoA 3.6 mm and 37 mL), while the novice showed higher variability (5.1 mm and 61 mL). Conclusion: AI-based measurements demonstrated repeatability comparable to experienced cardiologists. Despite moderate agreement and systematic differences in volumetric assessments, their reproducibility supports automated analysis for longitudinal echocardiographic monitoring.