Fig. 1: The model pipeline.
From: Improving model fairness in image-based computer-aided diagnosis
a We used four large-scale publicly available datasets (MIDRC, MIMIC-CXR, OHTS, and AREDS) with a diverse population to detect COVID-19 from CXR, thorax disease abnormality from CXR, primary open-angle glaucoma (POAG) from the optic disc, and late age-related macular degeneration (Late AMD) from color fundus photographs, respectively. b We trained a deep learning model with marginal ranking loss using the data specific to each disease. c We evaluated pairwise fairness across different subgroups, including sex, race, age, and genotypes, to determine if the model is equally fair for all individuals in each subgroup.
