Geographic Distribution of US Cohorts Used to Train Deep Learning Algorithms

This article investigates the geographic distribution of US cohorts utilized in training deep learning algorithms, particularly in medical imaging and diagnostics. It reveals a significant imbalance, with most cohorts originating from a limited number of states, potentially leading to algorithmic bias and reduced generalizability across diverse populations. The study emphasizes the critical need for more geographically diverse datasets to ensure fairness, accuracy, and equitable performance of AI in healthcare applications nationwide.