AI Algorithms Can Accurately Predict Patient Demographics From Chest Radigraphs

Background: Artificial intelligence (AI) algorithms are increasingly used to assist interpretation of imaging studies, for example, in the detection of acute intracranial hemorrhage and pulmonary embolism. There is a concern that, since AI deep learning models are trained on image databases, these databases may not reflect the diversity of the patient population. As a result, many AI programs may not accurately diagnose conditions involving underrepresented populations. Predetermination of a patient's population might ensure greater accuracy. Recent advances in deep learning technology have allowed researchers to see how AI algorithms classify images. Creating a visual heat map, which highlights regions of the image that influence a model’s decision, can increase confidence in an AI model and ensure the validity of decision making. In addition, these techniques can be used in an exploratory manner. Objective: To assess the ability of AI to classify the gender, age, ethnicity, a more...