{"id":"W4384070440","doi":"10.1148/ryai.220270","title":"The Subgroup Imperative: Chest Radiograph Classifier Generalization Gaps in Patient, Setting, and Pathology Subgroups","year":2023,"lang":"en","type":"article","venue":"Radiology Artificial Intelligence","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"ca_institutions":"Royal University Hospital; Vector Institute; York University; Kingston Health Sciences Centre; Queen's University; Trillium Health Centre; University of Toronto","funders":"","keywords":"Medicine; Chest radiograph; Classifier (UML); Subgroup analysis; Radiography; Generalization; Radiology; Pathology; Artificial intelligence; Meta-analysis","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001048297,0.0002412082,0.0003786684,0.0004330234,0.0003768218,0.0000493595,0.0001625322,0.0002864559,0.00005173837],"category_scores_gemma":[0.00114018,0.0001913593,0.00009028512,0.001234035,0.0008233119,0.00008798255,0.00009543844,0.0004201405,0.0001026273],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001162504,"about_ca_system_score_gemma":0.0001056477,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001033004,"about_ca_topic_score_gemma":0.0004856413,"domain_scores_codex":[0.9974771,0.0004854045,0.000654341,0.0006149606,0.000164731,0.0006034878],"domain_scores_gemma":[0.9979365,0.001278378,0.0001494103,0.0004034864,0.0001088563,0.0001233245],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001151427,0.0006189597,0.3693216,0.0002229925,0.0002403579,0.002424054,0.01787168,0.003528656,0.1490759,0.09397323,0.03562986,0.3259412],"study_design_scores_gemma":[0.0009128975,0.002750627,0.6796478,0.0003744694,0.0002591337,0.002027585,0.00569351,0.1075505,0.0886321,0.03963395,0.07091811,0.001599289],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.973538,0.001658723,0.00181233,0.02143782,0.0007148374,0.0006071829,0.0000090822,0.0001716635,0.00005035784],"genre_scores_gemma":[0.9911566,0.004010874,0.0002956137,0.003953333,0.0002394041,0.0001544665,0.00005862831,0.00003502662,0.00009603334],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.324342,"threshold_uncertainty_score":0.7803407,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03459874656786698,"score_gpt":0.3278431426661905,"score_spread":0.2932443960983235,"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."}}