Skip to content

Deep-learning AI Could Predict Heart Disease Morbidity From Chest X-ray

In This Article

  • Atherosclerotic cardiovascular disease (ASCVD) can lead to heart attack or stroke, but the current standard for calculating an ASCVD risk score involves analyzing a large variety of health information, which is not always available for all patients
  • Massachusetts General Hospital researchers are developing a deep-learning AI model that could predict ASCVD heart attack or stroke risk using only a chest X-ray
  • In a preliminary study, there was a significant association between the risk predicted by the deep learning model and observed major cardiac events
  • The results of the model were also largely similar to that of the traditional scoring method

Atherosclerotic cardiovascular disease (ASCVD) can lead to heart attack, stroke, or other potentially fatal conditions. The current standard for calculating an ASCVD risk score involves analyzing a large variety of health information, such as age, sex, race, systolic blood pressure, hypertension treatment, and more. Still, these variables are not always available to fully calculate the score.

Researchers at Massachusetts General Hospital are developing a new option for predicting the risk of death from ASCVD heart attack or stroke: a deep-learning AI model that searches chest X-ray images to identify patterns associated with ASCVD risk. Lead author Jakob Weiss, MD, a radiologist in the Cardiovascular Imaging Research Center at Mass General, and colleagues presented the results at the 2022 Radiological Society of North America annual meeting.

Developing the Model

The deep learning model was developed by 1) training the AI to examine chest X-rays, and 2) identifying indications of a major adverse cardiovascular event occurring within 10 years. The AI analyzed over 147,000 chest X-rays obtained from more than 40,000 participants in a separate cancer screening trial.

The researchers then tested the model by tasking it with estimating the 10-year heart disease risk scores of a separate cohort of 11,430 Mass General Brigham patients. None of the patients had ever experienced a major adverse cardiovascular event, but they were all potentially eligible for statin treatment.

Study Results

Of the Mass General Brigham patient cohort, 9.6% suffered a major adverse cardiac event (median follow-up of 10.3 years). There was a significant association between those observed major cardiac events and the risk predicted by the deep learning model.

The researchers also compared the AI's findings to the traditional ASCVD scoring method in 2,401 patients who had a chest X-ray and the necessary medical data on file. The results of the deep learning model were largely similar to that of the traditional scoring method.

While additional research is necessary to validate the model further, these preliminary results are promising. The model relies only on a single variable, suggesting that the approach could provide both diagnostic and prognostic information that ultimately improves the patient's course of treatment.

Learn about the Cardiovascular Imaging Research Center

Refer a patient to the Corrigan Minehan Heart Center

Related topics

Related

Researchers at Massachusetts General Hospital demonstrated that "CXR-Age," a convolutional neural network, can estimate biological age from a chest X-ray image. This biological age was better than chronological age at predicting longevity.

Related

James L. Januzzi, MD, and colleagues conducted a large randomized, placebo-controlled trial without any in-person visits, demonstrating canagliflozin was associated with a rapid and clinically meaningful improvement in the symptoms of heart failure regardless of ejection fraction and presence of diabetes.