Developing AI technology to determine the cause of left ventricular hypertrophy only with echocardiography...Accuracy up to 96%
Jun 02, 2025
A research team led by Yoon Yeon-yi, a professor of circulatory technology at Seoul National University Bundang Hospital, is drawing attention by developing artificial intelligence (AI) technology that accurately diagnoses left ventricular hypertrophy with only echocardiographic images and distinguishes causes.
The left ventricle is the core part of the heart that sends oxygenated blood from the lungs throughout the body, and plays an important role in supplying blood flow throughout the body. A condition in which the wall (myocardium) of the left ventricle is abnormally thickened and the heart function is deteriorated is called 'left ventricular hypertrophy', and it is caused by various causes such as hypertensive heart disease, hypertrophic cardiomyopathy, and heart amyloidosis. Since the treatment and prognosis vary depending on the causative disease, it is very important to accurately distinguish them.
Echocardiography is widely used as the primary test for diagnosis of left ventricular hypertrophy, but there is a limit to distinguishing minute structural differences in the ventricle with the naked eye of the examiner, so additional detailed examinations such as MRI are required. However, if the diagnosis is delayed in this process, the treatment may be delayed and it may lead to serious complications such as heart failure and sudden death, raising the need for a more efficient and reliable diagnosis method.
In response, Professor Yoon Yeon-yi's team conducted research to develop AI-based diagnostic technology that can identify the cause with only echocardiographic images.
The research team developed a model that can diagnose left ventricular hypertrophy and distinguish ▲ hypertensive cardiomyopathy ▲ hypertensive cardiomyopathy ▲ cardiomyopathy ▲ cardiomyopathy by digitizing a total of 19,839 characteristic information, including fine patterns and shape changes of myocardium in echocardiography images, so that AI can learn disease-specific patterns.
The performance of the AI model was evaluated using independent verification data from external hospitals, and the diagnostic accuracy was 96% for hypertrophic cardiomyopathy, 89% for cardiac amyloidosis, and 83% for hypertensive heart disease. This means that AI models can classify all three diseases with very high accuracy.
In particular, the diagnostic sensitivity of hypertensive heart disease was 33% in the conventional echocardiography method, but improved to 75% in the AI model. The F1 score for hypertrophic cardiomyopathy also increased from 0.57 to 0.87, indicating that the AI model overall performs better than the existing methods. Sensitivity is the rate of finding real patients without missing them, and the F1 score is a comprehensive indicator that evaluates the accuracy and consistency of the diagnosis together.
In addition, it is expected that the image parts judged importantly by AI during the analysis process will appear visually, and medical staff can check the evidence themselves, increasing the transparency and reliability of the diagnosis process, and that it will be highly likely to be used in actual clinical practice.
Professor Yun Yeon-yi said, `As the cause of left ventricular hypertrophy is delayed in clinical settings, treatment opportunities are often missed or the prognosis is poor.'"This study is of great significance in suggesting the possibility of using artificial intelligence to overcome the limitations of existing diagnosis and to evaluate the causative disease more quickly and objectively in the echocardiographic stage, the primary test." "We plan to expand our research to AI models that help discriminate between rare diseases such as Fabry's disease and Danone's disease, or physiological left ventricular hypertrophy in athletes," he added.
Meanwhile, the findings were published in the American Heart Association's prestigious journal 'Circulation: Cardiovascular Imaging'.
The left ventricle is the core part of the heart that sends oxygenated blood from the lungs throughout the body, and plays an important role in supplying blood flow throughout the body. A condition in which the wall (myocardium) of the left ventricle is abnormally thickened and the heart function is deteriorated is called 'left ventricular hypertrophy', and it is caused by various causes such as hypertensive heart disease, hypertrophic cardiomyopathy, and heart amyloidosis. Since the treatment and prognosis vary depending on the causative disease, it is very important to accurately distinguish them.
Echocardiography is widely used as the primary test for diagnosis of left ventricular hypertrophy, but there is a limit to distinguishing minute structural differences in the ventricle with the naked eye of the examiner, so additional detailed examinations such as MRI are required. However, if the diagnosis is delayed in this process, the treatment may be delayed and it may lead to serious complications such as heart failure and sudden death, raising the need for a more efficient and reliable diagnosis method.
In response, Professor Yoon Yeon-yi's team conducted research to develop AI-based diagnostic technology that can identify the cause with only echocardiographic images.
The research team developed a model that can diagnose left ventricular hypertrophy and distinguish ▲ hypertensive cardiomyopathy ▲ hypertensive cardiomyopathy ▲ cardiomyopathy ▲ cardiomyopathy by digitizing a total of 19,839 characteristic information, including fine patterns and shape changes of myocardium in echocardiography images, so that AI can learn disease-specific patterns.
The performance of the AI model was evaluated using independent verification data from external hospitals, and the diagnostic accuracy was 96% for hypertrophic cardiomyopathy, 89% for cardiac amyloidosis, and 83% for hypertensive heart disease. This means that AI models can classify all three diseases with very high accuracy.
In particular, the diagnostic sensitivity of hypertensive heart disease was 33% in the conventional echocardiography method, but improved to 75% in the AI model. The F1 score for hypertrophic cardiomyopathy also increased from 0.57 to 0.87, indicating that the AI model overall performs better than the existing methods. Sensitivity is the rate of finding real patients without missing them, and the F1 score is a comprehensive indicator that evaluates the accuracy and consistency of the diagnosis together.
In addition, it is expected that the image parts judged importantly by AI during the analysis process will appear visually, and medical staff can check the evidence themselves, increasing the transparency and reliability of the diagnosis process, and that it will be highly likely to be used in actual clinical practice.
Professor Yun Yeon-yi said, `As the cause of left ventricular hypertrophy is delayed in clinical settings, treatment opportunities are often missed or the prognosis is poor.'"This study is of great significance in suggesting the possibility of using artificial intelligence to overcome the limitations of existing diagnosis and to evaluate the causative disease more quickly and objectively in the echocardiographic stage, the primary test." "We plan to expand our research to AI models that help discriminate between rare diseases such as Fabry's disease and Danone's disease, or physiological left ventricular hypertrophy in athletes," he added.
Meanwhile, the findings were published in the American Heart Association's prestigious journal 'Circulation: Cardiovascular Imaging'.
|
This article was translated by Naver AI translator.
