A randomized crossover study of cardiologist reporting of severe aortic stenosis with/without assistance from...

Inaccurate echocardiographic diagnosis of severe aortic stenosis (AS) may be improved by AI. A randomised case-control fully-crossed, paired-reader, paired-case (MRMC) study evaluated the performance of AI in assisting cardiologist reporting of severe AS. 200 echos were independently adjudicated by expert cardiologists, 100 with severe AS (cases) and 100 without severe AS (controls). The AI was then applied and correctly classified all (100%) cases of severe AS, and 18 (36%) of moderate AS with signs of cardiac damage. Five independent cardiologists reported all echos according to the MRMC protocol. Without AI, cardiologists frequently misdiagnosed low-gradient severe AS. With AI, AS diagnosis rates were not improved, but AI improved reader concordance, referral for further investigation and reporting time. Cardiologists frequently misdiagnose low gradient severe AS, whereas measurement AI showed excellent accuracy. Lack of improvement in cardiologist accuracy with AI assistance highlights opportunities to improve trust in AI.