A routine heart ultrasound may contain hidden clues about a serious blood-flow problem that doctors normally need a more specialized test to measure.
Researchers at Mayo Clinic have developed an artificial intelligence system that can recognize those clues and identify people who may need further testing.
The technology is designed for patients with hypertrophic cardiomyopathy, or HCM. This is a genetic heart condition in which the muscle of the heart becomes thicker than normal.
HCM is not the same in every patient. Some people have few symptoms and may not know they have the condition, while others develop shortness of breath, chest discomfort, dizziness, fainting or difficulty exercising.
One reason symptoms can become severe is that the thickened heart muscle may block the route used by blood as it leaves the heart. Doctors call this left ventricular outflow tract obstruction.
The left ventricle is the heart’s main pumping chamber. With each heartbeat, it pushes oxygen-rich blood through the aorta and out to the rest of the body.
If the exit from this chamber becomes too narrow, blood cannot flow out as easily. The heart then has to work against the obstruction, which can contribute to symptoms and influence how doctors manage the disease.
About two-thirds of patients with HCM develop this type of obstruction. In some people it is present while they are resting, while in others it becomes clear only during exercise or when the heart is placed under additional stress.
Doctors usually investigate the problem with Doppler echocardiography. Like a standard ultrasound, the test uses sound waves, but Doppler also measures how fast blood is moving through different parts of the heart.
Fast blood flow through a narrowed area can help doctors estimate the severity of an obstruction. However, getting an accurate Doppler measurement depends on correctly positioning the ultrasound beam and requires appropriate training and experience.
This can create challenges in places without specialist heart-imaging services. Mayo Clinic researchers therefore wanted to know whether an AI system could identify likely obstruction using only the simpler ultrasound videos that are routinely collected.
These standard videos show the heart beating in real time but do not directly measure blood-flow speed. To a human observer, they may not provide enough information to confidently determine whether an important obstruction is present.
AI can approach medical images differently. Instead of looking only for the features doctors have traditionally been trained to recognize, a computer model can learn complex combinations of movement, shape and timing associated with a particular condition.
The study included a Mayo Clinic group of 1,833 patients. The researchers tested the model in 275 patients and then examined whether it would continue working in a separate group of 46 patients from a hospital in South Korea.
The AI used resting, non-Doppler ultrasound videos to predict whether a patient had a potentially important increase in pressure caused by obstruction to blood leaving the heart. The model performed best when it combined information from three standard ultrasound views.
Using several views makes sense because each angle reveals different parts of the heart. Combining them may allow the computer to build a more complete picture of how the thickened muscle affects movement and blood flow.
The system also showed an ability to flag some patients whose obstruction may become apparent mainly under stress. This could be useful because a normal-looking resting examination does not always rule out an important problem during physical activity.
Another encouraging result came from the South Korean patients. The model maintained strong performance even though these patients differed from the population used to develop the AI.
Testing an AI system at another hospital is important because medical algorithms can sometimes perform well only where they were created. Differences in patients, equipment and clinical practices can reduce accuracy when a model is moved to a new setting.
In part of the study, two expert echocardiographers were asked to assess the same ordinary ultrasound images without Doppler information. In some cases, the AI identified obstruction more accurately than the specialists.
This finding should not be interpreted as evidence that AI can replace cardiologists. Instead, it shows that routine images may contain subtle information that is extremely difficult for the human eye to recognize without the additional measurements provided by Doppler.
Senior author Dr. Imon Banerjee said the system is intended to complement existing testing. A positive AI result could prompt clinicians to perform Doppler measurements, stress testing or refer the patient to a center specializing in HCM.
The approach could eventually be useful with portable ultrasound equipment or in communities where advanced echocardiography is not readily available. A basic scan could potentially serve as an initial screening step, helping determine who needs more specialized evaluation.
The research was published in Circulation: Cardiovascular Imaging. It demonstrates how AI may add new information to a familiar medical test without requiring patients to undergo a completely new type of scan.
There are important reasons for caution, however. The independent South Korean validation included only 46 people, so the system still needs to be tested in much larger and more varied populations.
Researchers also need to know how the model performs with different ultrasound machines, image quality and clinical workflows. A tool that works well in a research study must still prove that it is reliable when used by many healthcare professionals in everyday practice.
Most importantly, future studies must determine whether AI-assisted screening actually helps patients. Earlier detection is valuable only if it leads to appropriate treatment, better symptoms, fewer complications or other meaningful improvements in care.
Overall, the study offers a promising example of AI finding useful signals in medical images that doctors already collect. If future trials confirm its accuracy and clinical value, the technology could help uncover hidden heart obstruction sooner and expand access to specialist-level screening in places where advanced testing is limited.
Source: Mayo Clinic.
