AI predicts long-term mortality by analyzing chest X-rays

Researchers from Massachusetts General Hospital have developed a convolutional neural network—called CXR-risk—to predict 12-year mortality from chest radiographs.

In a prognostic study of data from two randomized clinical trials, CXR-risk was able to predict long-term mortality independent of radiologists’ readings of the chest X-rays and other factors, such as age and smoking status. The neural network was able to do so without any other demographic or clinical information.

chest_6984-CROP.jpg
radiologist Matt Lungren, M.D., left, meets with graduate students, Jeremy Irvin and Pranav Rajpurkar to discuss the results of tests using the algorithm the students developed.

“Individuals at high risk of mortality based on chest radiography may benefit from prevention, screening and lifestyle interventions,” conclude the study’s authors.

According to the study, published in the journal JAMA Network Open, this was the first report of deep learning to predict long-term prognosis from chest radiographs.

Also See: Machine learning predicts 1-year mortality using EHR data

The CXR-risk score, classified as very low, low, moderate, high and very high, was based on chest X-rays for 16,000 patients from two earlier clinical trials.

The study showed that 53 percent of patients the convolutional neural network (CNN) identified as very high risk died over 12 years, while fewer than 4 percent of those that the deep learning model deemed as very low risk died.

“The CXR-risk score predicted multiple causes of death, including both lung cancer and non-cancer death due to cardiovascular and respiratory illness,” according to the study. “In fact, most deaths were from causes other than lung cancer. These observations suggest that this CNN should not be considered as a lung cancer detector. Instead, we speculate that it identified patterns on the chest radiograph not tied to a single diagnosis or disease but as a summary measure of underlying prognosis and health.”

“This is a new way to extract prognostic information from everyday diagnostic tests,” says co-author Michael Lu, MD, director of research for the MGH Division of Cardiovascular Imaging and assistant professor of radiology at Harvard Medical School. “It’s information that’s already there that we’re not using, that could improve people’s health.”

For reprint and licensing requests for this article, click here.