Researchers at Mayo Clinic have created an artificial intelligence tool capable of spotting signs of a significant heart blockage using standard ultrasound videos.
This innovation may assist in identifying patients who need more comprehensive cardiac testing, especially in clinics or hospitals lacking advanced ultrasound specialists.
The AI was primarily developed for individuals with hypertrophic cardiomyopathy, commonly called HCM. This inherited condition causes abnormal thickening of the heart muscle and can affect people of all ages.
While some with HCM experience few or no symptoms, others may find that the thickened heart muscle obstructs blood flow, leading to symptoms like shortness of breath, chest pain, fatigue, fainting, or irregular heart rhythms.
A key concern is an obstruction in the pathway where blood exits the heart’s main chamber, the left ventricle. This is known as the left ventricular outflow tract, or LVOT.
Approximately two-thirds of HCM patients develop some level of LVOT obstruction. When this pathway narrows, the heart must work harder to pump blood into the aorta, which can elevate symptoms and risk.
Detecting this obstruction is crucial for understanding patient symptoms and determining appropriate treatment, whether that involves medication or, in severe cases, specialized procedures.
Typically, clinicians measure LVOT obstruction through Doppler echocardiography, which uses sound waves to assess the speed and direction of blood flow within the heart.
While valuable, this technique demands skillful operation and interpretation by experienced operators. Accurate results depend heavily on proper positioning of the ultrasound probe and expert analysis.
Researchers asked if AI could analyze simpler, standard ultrasound videos—called B-mode images—that display the heart’s structures without Doppler data. These videos show the heart’s movement during each beat, capturing its shape and motion.
The AI was trained to recognize subtle patterns linked to significant obstruction, some of which are so slight that even seasoned specialists might struggle to identify them from basic images alone.
The study involved 1,833 Mayo Clinic patients. The model was tested on 275 of these patients and externally validated with 46 patients from a hospital in South Korea.
External testing is vital because it shows whether the AI can perform well across different patient populations and clinical settings, beyond the initial development environment.
The AI analyzed only resting-state ultrasound videos without Doppler data, aiming to predict which patients likely had a meaningful LVOT obstruction.
Results indicated the model was more accurate when combining information from three standard ultrasound views, rather than relying on just one. Multiple perspectives offered richer details about the heart’s structure and motion.
Interestingly, the AI also showed promise in detecting obstructions that become apparent mainly during physical stress or exercise, which is important since some patients may only develop significant blockage while active.
Despite differences between the U.S. and South Korean groups, the AI maintained strong performance, suggesting it might be adaptable across diverse populations, though larger studies are needed to confirm this.
In some cases, the AI identified obstructions more accurately than two expert echocardiographers reviewing the same non-Doppler images. While it doesn’t replace trained specialists, this highlights the difficulty of detecting subtle obstructions using standard images alone.
Lead researcher Dr. Imon Banerjee emphasized that this technology isn’t meant to replace Doppler ultrasound but could serve as an early warning system to flag patients who need further, detailed testing—such as stress tests or evaluation at specialized centers.
This could be especially helpful in settings where access to advanced cardiology imaging is limited. Portable ultrasound devices are increasingly common, but may lack the expertise for complex assessments.
If the AI reliably flags potentially problematic cases from basic videos, patients could be referred earlier for thorough evaluations, reducing the risk of missing critical obstructions.
The findings were published in the journal Circulation: Cardiovascular Imaging, a publication of the American Heart Association. They suggest that routine ultrasound images contain hidden information about blood flow obstruction that AI may be capable of detecting without explicit blood flow measurements.
While promising, the study has limitations. The external validation group was small, with only 46 patients, which doesn’t prove the system’s effectiveness across different hospitals, machines, or broader populations.
AI performance can also be affected by the quality of images and differences in patient characteristics compared to the training data. Larger, real-world studies are necessary to understand how this technology performs in typical clinical environments.
The team plans to expand validation efforts across diverse healthcare settings, different ultrasound devices, and larger patient groups. Future research will also assess how integrating this AI influences clinical decision-making and whether earlier detection improves patient outcomes.
In summary, this study highlights a valuable role for AI: uncovering hidden insights in tests already performed by doctors, rather than replacing established methods. With further validation, such tools could streamline early detection of heart obstructions, ensuring patients get the specialized care they need at the right time.
Source: Mayo Clinic.