Q&A: AI Predicts 10-Year Risk of Heart Attack and Stroke from X-Ray Images
In This Article
- Despite continued advances, there is still a need for a better way to identify patients at highest risk of cardiovascular disease, the most common cause of death
- Researchers at Massachusetts General Hospital have developed a deep learning convolutional neural network (CXR-CVD Risk) tool that predicts the 10-year risk of heart attack and stroke based on existing chest radiograph (X-ray) images
- This is important because many patients have existing chest radiographs but do not have the inputs an online risk calculator recommended by the American College of Cardiology/American Heart Association
In November, during the annual meeting of the Radiological Society of North America (RSNA), researchers from the Massachusetts General Hospital Cardiovascular Imaging Research Center (CIRC) presented their work using artificial intelligence (AI) to predict heart attack and stroke from chest radiograph (X-ray) images. This study was featured in the RSNA daily bulletin and on CNN. We spoke with co-authors Michael Lu, MD, MPH, co-director of CIRC and associate chair of Imaging Science, and Vineet Raghu, PhD, research faculty at the CIRC and instructor of Radiology at Harvard Medical School, about the study:
Subscribe to the latest updates from Radiology Advances in Motion
Q: What motivated you to do the study? What problem were you seeking to address?
Cardiovascular disease, including heart attack and stroke, is the most common cause of death. Effective tools to prevent cardiovascular disease include diet, exercise, and statin medications. However, cardiovascular prevention is underutilized, and there's a need for a better way to identify those at highest risk who would see the most benefit.
In the U.S., the 2019 American College of Cardiology/American Heart Association guidelines recommend determining statin eligibility using an online risk calculator called the Pooled Cohort Equations. This calculator requires a lipid panel, blood pressure, smoking information, and a history of diabetes/hypertension. This data is often unavailable in the electronic medical record, making it difficult to automate the identification of high-risk patients who should be on a statin.
We aimed to address this by developing a deep learning convolutional neural network (CXR-CVD Risk) tool that predicts the 10-year risk of heart attack and stroke based on existing chest radiograph (X-ray) images.
Q: What was the most important finding of the study?
Based on the pixels on a chest radiograph image, our CXR-CVD Risk model predicted a 10-year risk for cardiovascular events with similar performance to the current clinical standard, the Pooled Cohort Equation risk calculator.
This is important because chest radiographs are among the most common tests in medicine, especially in older adults. Many people have existing chest radiographs but not the inputs to the Pooled Cohort Equation risk calculator. Opportunistic screening of chest radiographs could identify additional patients who may benefit from statins and ultimately prevent heart attack and stroke.
Q: What was most novel about the study? What made it possible?
Chest radiographs are ordered to make a specific diagnosis, like pneumonia. The underlying idea of our research is that there is additional "hidden" information about aging and cardiovascular risk on the radiograph that can be extracted using AI. By showing a machine learning model of hundreds of thousands of chest radiographs, we can train it to predict long-term mortality, biological age, incident lung cancer, postop mortality, and coronary calcium score from chest radiographs and CT. In the last month, we published an MGB validation of our lung cancer risk model and a tool to triage chest pain using the initial ED chest radiograph. This work is made possible by 1) the availability of large databases of chest radiograph and CT images, and 2) advances in machine learning technology.
Q: Can you briefly describe the work of the Cardiovascular Imaging Research Center generally? How does the current study fit with the center's overall mission?
The Mass General Cardiovascular Imaging Research Center is a joint program between Radiology and Cardiology focused on using imaging to improve cardiovascular health. Much of CIRC's focus has been on conducting large multicenter clinical trials. Access to these large databases and our trial expertise has helped accelerate our machine learning program. We are excited to apply our first machine-learning tools in a clinical trial setting.
Authors on the RSNA abstract are Jakob Weiss, MD, Vineet Raghu, PhD, Kaavya Paruchuri, MD, Pradeep Natarajan, MD, MMSC, Hugo Aerts, PhD, and Michael T. Lu, MD, MPH. The work was partly supported by funding from the National Academy of Medicine and the American Heart Association.
Learn more about research in the Department of Radiology
Learn more about the Cardiovascular Imaging Research Center