Mayo’s AI Searches Routine Ultrasound for HCM Obstruction

Mayo Clinic researchers have built an AI model to flag potentially significant left ventricular outflow tract obstruction from resting, non-Doppler ultrasound videos in people with hypertrophic cardiomyopathy. The proposed role is not to replace the specialist measurement, but to identify patients who may need it sooner.

Reading a signal without Doppler

Hypertrophic cardiomyopathy, or HCM, causes the heart muscle to become abnormally thick. About two-thirds of patients develop left ventricular outflow tract, or LVOT, obstruction, which restricts blood leaving the heart and can cause chest pain or shortness of breath. Whether obstruction is present influences treatment and longer-term management.

The usual measurement relies on Doppler echocardiography, which requires precise alignment of the ultrasound beam and operator expertise. Mayo’s model instead takes resting B-mode, non-Doppler videos and predicts whether a patient has potentially significant obstruction. It is looking for patterns in routinely acquired images that investigators said may not be visible to the human eye.

A triage tool, not a diagnostic substitute

The model’s performance improved when it combined three standard ultrasound views. It also identified obstruction that may emerge only when the heart is under stress, according to the study. Those features make the system potentially useful as a prompt for a more complete workup rather than a final finding from a basic scan.

The intended clinical handoff

• Routine resting, non-Doppler ultrasound video supplies the model’s input.
• A prediction of potentially significant LVOT obstruction can flag a patient for confirmatory Doppler measurement, stress testing, or referral to an HCM specialty center.
• The researchers said this could be particularly relevant where comprehensive Doppler assessment or specialized echocardiography expertise is not readily available.

Encouraging results, narrow external evidence

In a subset of cases, the model identified obstruction more accurately than two expert echocardiographers reviewing the same non-Doppler images. The comparison underscores the difficulty of inferring LVOT obstruction from routine two-dimensional imaging alone.

The model maintained strong performance in the South Korean validation group despite differences from the development population, but that external group contained 46 patients. The authors call for prospective validation across clinical settings, ultrasound platforms, and patient populations. The work appeared in Circulation: Cardiovascular Imaging as “Beyond Doppler: Scalable AI Detection of LVOT Obstruction in HCM” and received no external funding.