Automated AI echo in ATTR cardiomyopathy: moderate expert agreement, repeatability comparable to cardiologists

Automated artificial intelligence echocardiography in 62 transthyretin cardiomyopathy patients (178 exams) showed moderate agreement with a human reader (intraclass correlation 0.65, 0.51) and repeatability similar to experienced cardiologists (limits of agreement 3.6 mm, 37 mL).

• Retrospective study of 62 patients with transthyretin cardiomyopathy and 178 serial annual echocardiograms compared automated AI (Us2.ai), a reference reader, a second cardiologist, and a novice.
• AI agreed moderately with the reference reader for wall thickness and left ventricular volume (ICC 0.65 and 0.51; biases -1.9 mm and -39 mL), while agreement between cardiologists was good (ICC 0.79 and 0.84; biases -0.2 mm and +3 mL).
• AI repeatability (limits of agreement 3.6 mm, 37 mL) was similar to experienced readers (3.0 mm/43 mL and 2.7 mm/44 mL) and better than a novice (5.1 mm/61 mL); expert oversight remains necessary.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.