{"id":"W4392672760","doi":"10.1016/j.ebiom.2024.105047","title":"Drop the shortcuts: image augmentation improves fairness and decreases AI detection of race and other demographics from medical images","year":2024,"lang":"en","type":"article","venue":"EBioMedicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Institute on Minority Health and Health Disparities; National Institute on Aging; National Institute of Biomedical Imaging and Bioengineering; Fogarty International Center; Canadian Institutes of Health Research; National Institutes of Health; Genentech; IXICO; H. Lundbeck A/S; Servier; Eisai; Gordon and Betty Moore Foundation; Northern California Institute for Research and Education; F. Hoffmann-La Roche; Radiological Society of North America; BioClinica; Pfizer; Biogen; National Heart, Lung, and Blood Institute; Novartis Pharmaceuticals Corporation; U.S. Department of Defense; Eli Lilly and Company; Bristol-Myers Squibb; Alzheimer's Disease Neuroimaging Initiative; Meso Scale Diagnostics; Alzheimer's Association; National Science Foundation; University of Southern California; National Science and Technology Council","keywords":"Demographics; Medicine; Magnetic resonance imaging; Artificial intelligence; Computer science; Machine learning; Demography; Radiology","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004696295,0.0001141351,0.0002015216,0.000168406,0.00008856147,0.00002920138,0.00004578039,0.0001079065,0.000129229],"category_scores_gemma":[0.0003729565,0.000070774,0.00003235232,0.0003328129,0.000575377,0.0001159677,0.00002327966,0.0002124447,0.000004137861],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002432536,"about_ca_system_score_gemma":0.0001205119,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00311888,"about_ca_topic_score_gemma":0.0002834901,"domain_scores_codex":[0.9988831,0.00005488708,0.00035271,0.0002392987,0.0003333437,0.0001366716],"domain_scores_gemma":[0.9990548,0.0004729085,0.0000546194,0.0001512429,0.0001037395,0.0001626999],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0001972752,0.00009567589,0.02715858,0.0005187058,0.0001309689,0.0000268911,0.004032237,6.068047e-8,0.3252483,0.0001430851,0.001442275,0.6410059],"study_design_scores_gemma":[0.0008857024,0.002482797,0.1846519,0.004125768,0.001479785,0.0005806332,0.03264909,0.01948933,0.720095,0.01175147,0.02126502,0.0005434327],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9607949,0.009468864,0.009812701,0.0189092,0.0005441211,0.0003281951,0.00001206748,0.00005816104,0.00007174395],"genre_scores_gemma":[0.9962285,0.001563888,0.0001589793,0.001274764,0.0006269242,0.00002445613,0.00002913576,0.00001745498,0.00007591848],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6404625,"threshold_uncertainty_score":0.4714834,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03583418101470353,"score_gpt":0.3995181496081336,"score_spread":0.3636839685934301,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}