New AI-based Approach Helps Identify At-Risk Patients from a Single X-Ray

Massachusetts General Hospital, Brigham and Women’s Hospital, and Harvard Medical School researchers have developed a deep learning model that can predict long-term mortality from a single chest X-ray. The new findings were recently published in the JAMA Network Open journal

The research has the potential to help identify patterns in chest radiographs not linked to a single diagnosis or disease but instead a summary of underlying health. 

“Even normal radiographs manifest additional minor abnormalities, such as aortic calcification or an enlarged heart, that may provide a new window into prognosis and longevity with the potential to inform decisions about lifestyle, screening, and prevention,” the researchers stated in their paper. “Physicians may interpret thousands of chest radiographs during a career, they rarely know the outcomes in these patients a decade later. Therefore, it is difficult to develop an intuition to articulate which features have long-term prognostic value.”

To test the proposed idea that a deep learning model can identify long-term mortality information from chest radiographs, the team developed a convolutional neural network named CXR-risk to predict 12-year mortality in patients. 

Using NVIDIA TITAN RTX GPUs with the cuDNN-accelerated PyTorch deep learning framework, the researchers trained their network on 41,000 images from a cohort of asymptomatic nonsmokers and smokers aged 55-74 enrolled at 10 different sites in the United States. 

For the classifier architecture and training, the researchers first chose to train the network with a staged classifier. During the first stage of training, the CNN was trained against a binary of death or incident cancer.  In the second stage of training, the team narrowed the classifier to death. 

According to the team, the model only used the chest radiograph images, excluding age, sex, risk factors, radiographic findings, and other factors. 

Gradient-Weighted Class Activation Maps (Grad-CAM) of Anatomy Contributing to the CXR-Risk Score:A and B, Grad-CAM (A) and chest radiograph (B) of a man in his 60s from the Prostate, Lung, Colorectal, and Ovarian (PLCO) trial who died of respiratory illness in 2 years. Grad-CAM highlights an enlarged heart with prominent pulmonary vasculature indicating pulmonary edema (very high-risk CXR-risk score). C and D, Grad-CAM (C) and chest radiograph (D) of a man in his 60s in the PLCO trial who died of cardiovascular illness in 7 years. Grad-CAM highlights the mediastinum and aortic knob, which may indicate cardiovascular health; sternotomy wires indicate previous cardiothoracic surgery (very high-risk CXR-risk score). E and F, Grad-CAM (E) and chest radiograph (F) of a man in his 60s in the National Lung Screening Trial who was alive at the end of 6-years follow-up. Grad-CAM highlights the extrathoracic soft-tissues, which may reflect body habitus (low-risk CXR-risk score). G and H, Grad-CAM (G) and chest radiograph (H) of a woman in her 50s in the PLCO trial who was alive at the end of 9-years follow-up. Grad-CAM highlights the shadow of the left breast and waist, which convey information about sex and habitus, important determinants of longevity (very low-risk CXR-risk score).

The findings of the CNN were complementary to the diagnosis of radiologists who took into account standard risk factors not seen during training, showing associations between lung cancer death, non-cancer cardiovascular death, and respiratory death. Most of the deaths were for causes other than lung cancer. 

“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,” the team stated. 

The team says that further research is necessary to determine how this study can improve individual and population health. 

Read more>