AI model helps clinicians detect heart obstruction using routine ultrasound images

Mayo Clinic scientists have created and verified an artificial intelligence (AI) model that can detect a potentially significant heart obstruction using standard ultrasound videos, bypassing the need for specialized Doppler imaging. This breakthrough technology could assist clinicians in identifying patients with hypertrophic cardiomyopathy (HCM) who may require further testing, particularly in areas where echocardiography expertise is scarce.

Hypertrophic cardiomyopathy is a genetic disorder that thickens the heart muscle, with around two-thirds of patients developing left ventricular outflow tract (LVOT) obstruction, leading to symptoms like chest pain and shortness of breath during exercise or when lying flat. Early detection of LVOT obstruction is crucial for guiding treatment decisions and long-term management.

Doppler echocardiography, typically used to measure LVOT obstruction, relies on precise ultrasound-beam alignment and operator expertise. The researchers aimed to determine if AI could identify subtle patterns in routinely obtained B-mode ultrasound videos that are difficult for humans to discern, to detect LVOT obstruction earlier and enable timely confirmatory Doppler evaluation.

The study involved 1,833 patients from the Mayo Clinic cohort, with the model tested on 275 patients and externally validated on 46 patients from a hospital in South Korea. The AI model analyzed resting, non-Doppler ultrasound videos to predict the presence of a potentially significant obstruction to blood leaving the heart. The study found that integrating data from three standard ultrasound views enhanced the model's ability to distinguish patients with elevated LVOT gradients and identify obstruction that may only appear under stress.

Notably, the model outperformed two expert echocardiographers in identifying obstruction from the same non-Doppler images in a subset of cases. While the AI model should complement, not replace, Doppler echocardiography, it could help early detection and prompt confirmatory Doppler measurements, stress testing, or referral to an HCM specialty center.

The technology could also facilitate evaluation using portable ultrasound or in resource-limited settings, expanding access to earlier screening and risk assessment. The researchers plan to conduct additional prospective validation across a wider range of clinical settings, ultrasound platforms, and patient populations.