RadAI

This data was collected in a post-deployment setting of DrAidTM, an AI-supported chest X-ray interpretation tool. DrAidTM employs an AI model that takes in a chest X-ray image and outputs the presence or absence of 21 pathologies and findings. In the interface, the radiologist is presented with (1) the patient’s medical and Rx history, (2) a chest X-ray in DICOM format, (3) relevant demographics such as age and gender, (4) predicted pathologies and other findings, and (5) regions of interest (ROIs) associated with each AI-predicted finding. The user can view ROIs, add AI findings to a report, request a second opinion, search for additional pathologies to add, and generate both internal and patient-accessible reports.
Data consists of predicted pathologies, final radiologist report inclusions, and demographics for 10,569 patients. Patients were randomly sampled from those whose X-rays were interpreted using DrAidTM at Nam Dinh Hospital in Vietnam between October of 2019 and February of 2022. Further, in order to ensure high-quality samples, only images interpreted by a radiologist whose familiarity with the DrAidTM tool was in the top-10, as measured by number of images interpreted using the tool, were included in the dataset. Pathologies and findings detected include cardiomegaly, fracture, lung lesion, pleural effusion, pneumothorax, atelectasis, consolidation, pneumonia, edema, cavitation, fibrosis, enlarged cardiomediastinum, widened mediastinum, pleural other, medical device, COVID-19, mass, nodule, mass or nodule (unknown), lung opacity, and tuberculosis.