Chest X-ray is the most commonly conducted imaging exam in medicine. According to a study by researchers at Massachusetts General Hospital (MGH), it holds “hidden” prognostic information that can be extracted using artificial intelligence (AI).
The research findings have been published in the JAMA Network Open on July 19th, 2019, and could help to recognize patients most likely to gain from screening and preventive medicine for lung cancer, heart disease, and other conditions.
AI technology automates several facets of everyday life, such as photo tagging on social media, the smartphone’s speech-recognition function, and self-driving cars. AI is also accountable for key advances in medicine; for instance, a number of groups have applied AI to automate the diagnosis of chest X-rays for spotting signs of tuberculosis and pneumonia.
If this technology can make diagnoses, asked radiologist Michael Lu, MD, MPH, could it also recognize people at high risk of future lung cancer, heart attack, or death? Lu, who is the director of research for the MGH Division of Cardiovascular Imaging and assistant professor of Radiology at Harvard Medical School, and his colleagues created a convolutional neural network, an advanced AI tool for examining visual information, called CXR-risk.
CXR-risk was taught by having the network examine over 85,000 chest X-rays from 42,000 subjects who participated in a previous clinical trial. Each image was matched with crucial data: Did the person expire during a 12-year period? The aim was for CXR-risk to absorb the features or combinations of features on a chest X-ray image that perfectly predict health and mortality.
Subsequently, Lu and colleagues examined CXR-risk using chest X-rays for 16,000 patients from two previous clinical trials. They learned that 53% of people in the neural network identified as “very high risk” expired over 12 years, compared to lesser than 4% of those that CXR-risk categorized as “very low risk.”
As part of the research, it was found that CXR-risk provided information that foretells long-term mortality, autonomous of radiologists’ readings of the X-rays and other parameters, such as smoking status and age.
Lu trusts this new tool will be even more precise when integrated with other risk parameters, such as smoking status and genetics. Early identification of at-risk patients could offer more to preventive and treatment programs.
This is a new way to extract prognostic information from everyday diagnostic tests. It’s information that’s already there that we’re not using, that could improve people’s health.
Michael Lu, MD, MPH, Radiologist and Director of Research, Division of Cardiovascular Imaging, MGH