AI Predicts Heart Attack

Early research shows promise in AI for predicting the risk of stroke or heart attack using one chest X-ray

Early research indicates that artificial intelligence could be used to predict the 10-year risk of dying from a stroke or heart attack from one chest X-ray.

On Tuesday, the preliminary findings were presented at the annual meeting of the Radiological Society of North America. The final draft of the research is still in progress and has not yet been published in a medical journal.

To train an artificial intelligence program, researchers used almost 150,000 chest radiographs to identify patterns in images that could indicate the risk of major cardiovascular disease events. The program was also tested on an additional group of approximately 11,000 people. They found a “significant association” between AI-predicted risk levels and actual occurrences of major cardiovascular diseases.

The ASCVD risk score is the clinical standard for analyzing cardiovascular risk. It weighs several patient data points and is highly associated with adverse cardiovascular events such as high blood pressure or smoking history.

For people at a 10-year high risk of developing a serious condition, statin medication may be recommended. Early findings show that the AI model works the same way as the traditional risk calculator.

“X-rays have been recognized for their ability to capture more information than traditional diagnostic findings,” stated Dr. Jakob Weiss who is the lead researcher and a radiologist at Massachusetts General Hospital and Harvard Medical School’s AI in Medicine program.

He said that sometimes the AI findings are in line with traditional radiology readings, while other times it may pick up on things that might have been missed.

“A part of it is anatomical changes that we would also detect with our naked eyes and that make physiological sense. Let’s suppose there is increased blood pressure or cardiac disease. These are also findings that can be seen on a regular chest radiograph. Weiss stated that although we believe a lot of information is captured or extracted from the scan, it’s not possible to make sense of it as a traditional radiologist.

He said that it has a black box character, which can make it difficult to communicate risk to patients without a clear explanation.

Donald Lloyd-Jones is a former president of the American Heart Association and Northwestern University’s Feinberg School of Medicine chair of preventive medicine. He was also a co-chair of the risk assessment panel that created the ASCVD risk calculator in 2013. In 2018, he was a key player when the guidelines were revised to emphasize the link between personal medical history and the risk score.

Although he was not part of the new AI research, he believes it is important to continue the field’s progress.

He said, “This is precisely the type of application that artificial Intelligence is best for.” “So, we must continue doing things like this to truly understand if it is possible to find, especially, patients who would otherwise fall through the cracks. This is where I believe it might be most helpful.”

However, it is important to collect all patient data points that are used in the risk calculator. They can be used for action. No matter whether the risk calculation is based on an AI model or a statistical formula, personalized patient assessments will be key to achieving the best results.

“We don’t cure smoking with a chest X-ray. Lloyd-Jones stated that we need to work with patients to find ways to quit smoking. The risk calculator is only one aspect of risk assessment. It’s not the sole part. This involves both the doctor and patient in a conversation about the patient’s risk and what we believe a statin might do for them.

Weiss and his co-authors used chest X-rays taken from participants in the National Cancer Institute’s Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial to train the AI. The AI was tested on patients who underwent routine outpatient chest radiographs at Mass General Brigham. They were eligible for statin therapy and had an average age of 60.

To validate the deep-learning model, additional research is required, including a controlled, randomized trial.

About The Author

Leave a Comment

Scroll to Top