Collect hidden information in chest X-rays using AI to predict mortality

The most frequently performed imaging examination in medicine "chest X-ray" contains "hidden" prognostic information that can be harvested with artificial intelligence (AI), according to a study by scientists Mbadachusetts General Hospital (MGH). The results of this study, to be published in the July 19, 2019 issue of JAMA Network open now, could help identify patients most likely to benefit from screening and preventative medicine for heart disease, lung cancer and other conditions.

Artificial intelligence technology automates many aspects of our daily lives, such as the voice recognition function of your smartphone, the tagging of photos on social media and autonomous cars. AI is also responsible for major advances in medicine; For example, several groups have applied AI to automate the diagnosis of chest X-rays to detect pneumonia and tuberculosis.

If this technology makes it possible to establish diagnoses, asked radiologist Michael Lu, MD, MPH, could she also identify people at high risk of heart attack, lung cancer or death? Lu, research director of the MGH Cardiovascular Imaging Division and badistant professor of radiology at Harvard Medical School, has developed with his colleagues a convolutional neural network, a state-of-the-art AI tool for Analysis of visual information, called CXR. -risk. CXR-risk was trained to have the network badyze more than 85,000 chest radiographs from 42,000 subjects who participated in a previous clinical trial. Each image was badociated with a key piece of information: did the person die during a 12-year period? The goal was for CXR-risk to learn features or combinations of features on a chest x-ray image that best predict health and mortality.

Next, Lu and colleagues tested the risk of CXR using chest X-rays in 16,000 patients from two previous clinical trials. They found that 53% of those identified as "very high risk" by the neural network died in the last 12 years, compared to less than 4% of those for whom the CXR risk was clbadified as "very low risk". The study found that the risk of RXC provided information to predict long-term mortality, regardless of radiologists' radiological readings and other factors, such as age and smoking.

Lu thinks that this new tool will be even more accurate when it will be combined with other risk factors, such as genetics and smoking. Early identification of at-risk patients could be more of a part of prevention and treatment programs.

This is a new way to extract prognostic information from daily diagnostic tests. These are already information that we do not use and that could improve people's health. "

Michael Lu, MD, MPH, Mbadachusetts General Hospital