Artificial intelligence-estimated “Chest X-ray Age” among atomic bomb survivors (Part1 Application to AHS and characterization of the features)

Summary

In this study on AHS participants, we aim to apply an artificial intelligence convolutional neural network model recently developed in US to predict “Chest X-ray Age (CXR-Age)” from CXR images. We evaluate their model performance on AHS participants and improve the model though transfer learning. We also aim to characterize the features contributing to CXR-Age. Based on this study, we will subsequently propose an RP to evaluate the association among 1) radiation exposure, 2) CXR-Age and features extracted with the model, and 3) overall mortality and disease-specific incidence/mortality.