Researchers at Mayo Clinic have developed an artificial intelligence tool that can detect signs of an important heart obstruction using ordinary ultrasound videos.
The technology may help identify patients who need more detailed heart testing, especially in hospitals or clinics without highly specialized ultrasound expertise.
The AI system was designed for people with hypertrophic cardiomyopathy, commonly called HCM. This inherited heart condition causes the heart muscle to become unusually thick and can affect people of many ages.
In some people, HCM causes few or no symptoms. In others, the thickened heart muscle can interfere with the heart’s ability to fill or pump blood and may lead to shortness of breath, chest pain, tiredness, fainting or abnormal heart rhythms.
One particularly important problem is an obstruction in the pathway where blood leaves the heart’s main pumping chamber, the left ventricle. This area is known as the left ventricular outflow tract, or LVOT.
About two-thirds of people with HCM can develop some degree of LVOT obstruction. When the pathway becomes too narrow, the heart must work harder to push blood into the body’s main artery.
Identifying this obstruction matters because it can help doctors understand a patient’s symptoms and choose appropriate treatment. Some patients may need medicines, while others with severe disease may require specialized procedures or other therapies.
Doctors usually measure LVOT obstruction with a type of heart ultrasound called Doppler echocardiography. Doppler uses sound waves to measure the speed and direction of blood moving through the heart.
This test can provide valuable information, but getting an accurate measurement requires skill. The ultrasound beam must be carefully positioned, and the person performing or interpreting the examination needs experience with HCM.
Mayo Clinic researchers asked whether AI could find clues to obstruction in simpler ultrasound videos that do not contain Doppler measurements. These standard black-and-white moving images, sometimes called B-mode ultrasound, show the heart’s structures as they move with each beat.
The researchers trained the computer model to recognize patterns associated with significant obstruction. Some of these patterns may be so subtle that even experienced specialists cannot reliably identify them from ordinary ultrasound images alone.
The study included 1,833 patients in the Mayo Clinic group. The researchers tested the model in 275 patients and then performed an external validation using 46 patients treated at a hospital in South Korea.
External validation is an important step in medical AI research. A model can appear highly accurate when tested on patients similar to those used during development, but its real value depends on whether it can also work in people from other hospitals and populations.
The AI analyzed only resting ultrasound videos without Doppler information. Its goal was to predict whether a patient was likely to have a potentially important obstruction to blood leaving the left ventricle.
The researchers found that the model performed better when it combined information from three common ultrasound views rather than relying on a single view. Looking at the heart from several angles gave the AI more information about its shape and movement.
The model also showed potential for detecting patients whose obstruction becomes noticeable mainly when the heart is placed under stress. This is important because some people may have little obstruction while resting but develop a significant problem during exercise or other physical stress.
The AI continued to perform strongly in the South Korean patients despite differences between that group and the population used to develop the system. This result suggests the approach may have the ability to work across different populations, although much larger studies are still needed.
In a subset of cases, the AI was more accurate than two expert echocardiographers who reviewed the same standard non-Doppler images. This does not mean the computer is a better heart specialist, but it demonstrates how difficult this particular problem can be without Doppler measurements.
Senior author Dr. Imon Banerjee emphasized that the technology is not intended to replace Doppler echocardiography. Instead, it could act as an early warning system that identifies patients who should receive a full Doppler examination, stress testing or assessment at an HCM specialty center.
This could be particularly useful in places where comprehensive heart ultrasound services are difficult to access. Portable ultrasound machines are becoming more common, but they may be used in settings where advanced imaging expertise is not immediately available.
If the AI can reliably flag suspicious cases from basic ultrasound videos, patients could potentially be referred for more complete testing sooner. Earlier recognition may reduce the chance that an important obstruction is overlooked.
The study findings were published in Circulation: Cardiovascular Imaging, a journal of the American Heart Association. The work provides evidence that routine ultrasound images contain information about blood-flow obstruction that AI may be able to extract even without directly measuring blood flow.
The results are promising, but the study also has clear limitations. The external validation group contained only 46 patients, which is far too small to prove that the system will perform equally well across different countries, ultrasound machines and clinical environments.
AI systems can also be affected by image quality and by differences between the patients used for training and those encountered in everyday care. Prospective studies, in which the technology is tested in real clinical practice, will therefore be important.
The researchers plan to validate the model across more healthcare settings, ultrasound platforms and patient populations. They will also need to determine how using the AI changes real clinical decisions and whether earlier detection ultimately improves patient outcomes.
Overall, the study shows a useful role for medical AI: finding hidden information in tests doctors already perform rather than replacing established diagnostic methods. If larger studies confirm the results, the technology could become a practical screening aid that helps more people with HCM receive specialized testing at the right time.
Source: Mayo Clinic.
