A research team at Seoul National University Bundang Hospital (SNUBH) has developed an artificial intelligence (AI) technology that can diagnose left ventricular hypertrophy (LVH) and identify its underlying causes using only echocardiographic images.

The hospital expects that the breakthrough will potentially enhance early and accurate cardiac diagnosis without requiring additional imaging such as MRI.

The left ventricle is the heart’s main chamber for pumping oxygen-rich blood throughout the body. LVH refers to abnormal thickening of the ventricular wall, which can lead to impaired cardiac function. It can result from various conditions, including hypertensive heart disease, hypertrophic cardiomyopathy, and cardiac amyloidosis. Since treatments and prognoses vary depending on the cause, accurate differentiation is critical in clinical practice.

Although echocardiography is widely used as the first-line test for diagnosing LVH, its effectiveness has been limited by the difficulty in visually identifying subtle myocardial differences. More detailed imaging, such as cardiac MRI, is often required -- potentially delaying treatment and increasing the risk of complications such as heart failure or sudden cardiac death.

To address these limitations, the team, led by Professor Yoon Yeon-yee of the Division of Cardiology, developed an AI model trained on 19,839 quantified features extracted from echocardiographic videos, capturing fine patterns and morphological changes in the myocardium.

The AI was designed to detect the presence of LVH and differentiate among its three most common causes -- hypertensive heart disease, hypertrophic cardiomyopathy, and cardiac amyloidosis.

When validated using independent datasets from external hospitals, the AI model demonstrated high diagnostic accuracy of 96 percent for hypertrophic cardiomyopathy, 89 percent for cardiac amyloidosis, and 83 percent for hypertensive heart disease.

Notably, the sensitivity for hypertensive heart disease improved from 33 percent using conventional echocardiography to 75 percent with the AI model. The F1 score for hypertrophic cardiomyopathy, a combined metric of precision and recall, also increased from 0.57 to 0.87, underscoring the model’s superior diagnostic performance.

The system also visualizes which parts of the ultrasound image contributed most to its decision-making process, providing transparency and supporting clinical confidence in its use.

“Delayed identification of the underlying cause of LVH in clinical settings often leads to missed treatment opportunities and poorer outcomes,” Professor Yoon said. “This study highlights how AI can overcome the limitations of conventional diagnosis and enable faster, more objective assessment at the echocardiography stage.”

The research will be expanded to include rare diseases such as Fabry disease and Danon disease, as well as physiological LVH observed in athletes, Yoon added.

The study was published in Circulation: Cardiovascular Imaging.