{"id":"W4409259931","doi":"10.1371/journal.pdig.0000811","title":"Subgroup evaluation to understand performance gaps in deep learning-based classification of regions of interest on mammography","year":2025,"lang":"en","type":"article","venue":"PLOS Digital Health","topic":"AI in cancer detection","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"National Cancer Institute","keywords":"False positive paradox; Medicine; Mammography; Univariate; Artificial intelligence; Multivariate statistics; Subgroup analysis; Convolutional neural network; Breast cancer; Internal medicine; Computer science; Machine learning; Confidence interval; Cancer","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":[0.0004217987,0.00008369314,0.0001554106,0.0005832629,0.00005636086,0.00003490639,0.0002501711,0.00003991961,0.000001384763],"category_scores_gemma":[0.00009118502,0.00008808451,0.00003702501,0.001314757,0.00004380137,0.0003809528,0.00004049676,0.0001349115,0.000003178889],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004737299,"about_ca_system_score_gemma":0.000279601,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002897777,"about_ca_topic_score_gemma":0.0002979174,"domain_scores_codex":[0.9988263,0.00008986925,0.0003628111,0.0002651888,0.0002805761,0.0001752254],"domain_scores_gemma":[0.9991397,0.0001307897,0.0002073012,0.0003248612,0.0001447964,0.00005260401],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006191885,0.001629796,0.1008388,0.00196428,0.00005444963,6.563005e-7,0.003998563,0.04031823,0.0007397179,0.06437298,0.0003674171,0.7850959],"study_design_scores_gemma":[0.00107184,0.00268516,0.2693689,0.001541038,0.000007906147,6.261854e-7,0.0006019221,0.7141212,0.00504507,0.00519397,0.000167967,0.0001943593],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7427968,0.00008462475,0.2532535,0.001695094,0.0001231687,0.0005092895,0.000003203517,0.0000485509,0.001485788],"genre_scores_gemma":[0.9992992,0.00001879224,0.0004723204,0.0001504163,0.000006700976,0.00002601914,0.000007763751,0.000005260762,0.0000135674],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7849016,"threshold_uncertainty_score":0.3591982,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09049813221488646,"score_gpt":0.3237286303025327,"score_spread":0.2332304980876462,"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."}}