AI Could Predict 10-Year Risk of Heart Disease With a Chest X-Ray, Early Study Shows

In a promising early study, an artificial intelligence (AI) model could predict patients’ 10-year risk of death from a heart attack or stroke using a single chest X-ray.

Traditionally, health providers use an ASCVD risk estimator to predict a patient’s 10-year risk of atherosclerosis—a build-up of cholesterol and fats in the artery walls. This approach often requires a cardiologist to take a patient’s blood pressure and run various tab tests.

A team of researchers now says an advanced AI model can use chest X-ray images to predict an individual’s cardiovascular death risk over 10 years with similar accuracy to the traditional risk estimator. Chest X-rays are already common for screening for many maladies. If an AI model can take advantage of this popular imaging tool, it can help identify patients at high risk for heart disease who might not have otherwise visited a cardiologist.

Such patients could take a statin or blood pressure medication to decrease their odds of suffering a heart attack or stroke, said the study’s lead author, Jakob Weiss, MD, a radiologist affiliated with the Cardiovascular Imaging Research Center at Massachusetts General Hospital and the AI in Medicine program at the Brigham and Women’s Hospital in Boston.

The work was presented at a meeting of the Radiological Society of North America on November 30.

Weiss said the predictive model is not meant to replace the traditional risk score calculator. But if it’s given the green light, the model could be used to predict health outcomes for people who might otherwise go unnoticed.

“With our model, we would be able to identify these patients and say, ‘you are at an increased risk to develop stroke or heart attack within the next 10 years. Please go see your cardiologist and check whether you qualify to get a statin, for example, or blood pressure medication to reduce your risk,’” Weiss said.

Training a Computer to Predict Heart Disease

Deep learning is a complex type of AI. For this study, the researchers trained a deep learning model to look for cardiovascular event risk by feeding it more than 147,000 chest X-rays from more than 40,000 people and told the computer which of those patients died of heart disease over 10 years. The data came from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial, a multi-center, randomized controlled trial designed by the National Cancer Institute.

With this approach, scientists give the machine a set of rules at the beginning and a set of outcomes at the end, and the machine “will sort of work out the middle steps,” said Alan Kwan, MD, a cardiologist and cardiac imaging physician scientist at Cedars-Sinai Medical Center.

The idea is that the computer can find certain markers of heart issues that may be invisible to cardiologists.

“Because this is sort of done mathematically and not in a way that I think is easy for humans to comprehend, some people have had concerns about artificial intelligence being a black box,” Kwan told Verywell. “You put something in, and you get something out... but you don’t really understand what’s happening in the middle.”

To test that the model works on data it’s never seen before, researchers fed the model additional images from a separate group of 11,430 patients who had a routine chest X-ray at Mass General Brigham hospital.

There was a significant correlation between the AI model’s risk predictions and the actual outcomes for the nearly 10% of patients who experienced a major adverse cardiac event at some point in the following 10.3 years.

A fifth of the patients had enough data in their medical records to calculate their 10-year risk of cardiovascular disease-related death using the ASCVD approach. The traditional method and deep learning model performed similarly in predicting the 10-year cardiovascular disease-related death risks for patients

Kwan, who isn’t involved with the study, said the results were promising, but the model may need to become more precise before it can be put into clinical practice.

Weiss said he expected the model to be tested in a more diverse group of patients in a controlled, randomized trial.

How AI May Fit Into the Future of Health Care

There are various ethical considerations when using artificial intelligence tools in medicine. Computer models can contain biases if they’re not trained with diverse data. The medical community must also consider how to safely implement AI in health systems.

Last week, a group called the Coalition for Health AI shared a plan to address some of these questions and ensure that AI health models perform safely and accurately.

Despite the unknowns, Kwan said AI will continue to become an important tool for screening and diagnostics.

“Computers’ ability to take in large amounts of information, process it, and put out something meaningful has now sort of surpassed what people are able to do. Leveraging that only makes sense if we want to progress in our care for patients,” Kwan said. “Does that mean that these predictive models are going to replace physicians? I think it’s unlikely. It represents sort of a tool in a physician’s armamentarium for how they manage patients.”

X-rays are two-dimensional images that are easy to process with artificial intelligence. As the technology progresses, Weiss said scientists may be able to study three-dimensional cross-sectional images, like those from CT and MRI scans.

Machine learning has shown promise for early detection of lung disease and various cancers. In a future where artificial intelligence is incorporated into patient care, Weiss said it’s possible that clinicians could find a predictive score for various conditions with a single X-ray.

“We weren’t able to extract this information because we didn’t have the tools and technology. I think we’re now at the point where we can do this, and maybe the field of radiology will really start changing from a primarily diagnostic and subjective field to a more prognostic and objective field,” Weiss said.