AI model flags possible heart obstruction in HCM from routine ultrasound

Three standard ultrasound views gave a Mayo Clinic AI model enough information to flag a potentially significant obstruction in the heart, without using specialized Doppler imaging. The model was tested in 275 patients and externally validated in 46 patients at a hospital in South Korea, using only resting, non-Doppler ultrasound videos.

The target is left ventricular outflow tract obstruction, or LVOT obstruction: a narrowing that restricts blood leaving the heart. It matters in hypertrophic cardiomyopathy, a genetic condition that thickens the heart muscle; about two-thirds of patients with HCM develop the obstruction. Identifying it can affect treatment decisions and long-term follow-up.

Doppler echocardiography measures blood flow, but it depends on precise alignment between the ultrasound beam and the heart, as well as operator expertise. The Mayo team instead trained its model to recognize subtle patterns in ordinary two-dimensional videos. Combining three standard views improved its ability to distinguish patients with elevated LVOT gradients, including obstruction that may appear only when the heart is under stress.

So what changes in practice? The model could help clinicians flag HCM patients who need confirmatory Doppler measurements, stress testing or referral to an HCM specialty center. That is particularly relevant in settings with limited echocardiography expertise, and could support screening with portable ultrasound; the system is intended to guide further evaluation, not replace Doppler echocardiography.

The results also come with a clear boundary. The model maintained strong performance in the South Korean group despite differences from the Mayo patients, and in a subset of cases it outperformed two expert echocardiographers on the same non-Doppler images. But the researchers are calling for prospective validation across broader clinical settings, ultrasound platforms and patient populations before its clinical role is established.