{"id":"W4414959082","doi":"10.3390/diagnostics15192536","title":"Performance of ChatGPT-4o in Determining Radiology–Pathology Concordance and Management Recommendations Following Image-Guided Breast Biopsies","year":2025,"lang":"en","type":"article","venue":"Diagnostics","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Concordance; Gold standard (test); Biopsy; Conservative management; Breast imaging; Multidisciplinary approach; Retrospective cohort study","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003659174,0.0001079672,0.0003172786,0.0002429516,0.00005938249,0.00000956934,0.0000793388,0.00005538481,0.000007624511],"category_scores_gemma":[0.0007275824,0.0001049262,0.0000383598,0.0002277622,0.0001655373,0.00005243503,0.00009307631,0.0002179997,0.000001550566],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004729769,"about_ca_system_score_gemma":0.00003234212,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002481463,"about_ca_topic_score_gemma":0.000004385837,"domain_scores_codex":[0.9991364,0.00004834016,0.0003385755,0.0002039934,0.00007097671,0.0002017303],"domain_scores_gemma":[0.9992122,0.000458147,0.00007861738,0.0001677506,0.00003585636,0.00004742038],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002054955,0.00005838674,0.9185423,0.0002745755,0.00005226633,0.00024388,0.0001958553,0.00002680815,0.0008433249,0.0005527365,0.001100946,0.07808836],"study_design_scores_gemma":[0.002559698,0.0001282422,0.9586663,0.001877608,0.000167331,0.000413781,0.0003621124,0.03355299,0.0004064238,0.00008660736,0.00162814,0.0001507929],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9914974,0.000517486,0.0009637437,0.003849213,0.0003810932,0.0002183489,0.000005027119,0.00002449297,0.002543168],"genre_scores_gemma":[0.9792516,0.005264362,0.01435794,0.0008966643,0.00003121595,0.00002944981,0.00002502764,0.0000108022,0.0001329854],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07793757,"threshold_uncertainty_score":0.4278765,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009463016648554454,"score_gpt":0.3088086472987807,"score_spread":0.2993456306502262,"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."}}