{"id":"W4412163704","doi":"10.1158/1557-3265.aimachine-b028","title":"Abstract B028: Assessing variability in training deep models in delineating Prostate gland anatomy between clinical experts","year":2025,"lang":"en","type":"article","venue":"Clinical Cancer Research","topic":"Medical Imaging and Analysis","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Prostate gland; Prostate cancer; Prostate; Medicine; Anatomy; Pathology; Biology; Medical physics; Cancer; Internal medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.02461123,0.0001982073,0.0008516459,0.000415992,0.0001015305,0.0001511453,0.0004256667,0.0003374797,0.00008575457],"category_scores_gemma":[0.00639613,0.0001824305,0.0001851844,0.001426505,0.0003575729,0.000294817,0.0001591093,0.002688114,0.000005330695],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002802212,"about_ca_system_score_gemma":0.0005195618,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007900143,"about_ca_topic_score_gemma":0.0005223993,"domain_scores_codex":[0.9944915,0.001223268,0.002100136,0.0007119238,0.0006028909,0.0008703073],"domain_scores_gemma":[0.9910685,0.007911726,0.00006660355,0.000445237,0.0001878274,0.0003201072],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001429522,0.00008092738,0.5607542,0.0001313507,0.00004705671,0.00003265193,0.0002748422,0.01025901,0.00001457775,0.00002001664,0.0001981437,0.4281729],"study_design_scores_gemma":[0.001450881,0.00001653328,0.2419671,0.0008612195,0.00001903424,2.128895e-7,0.000626457,0.7501236,0.00002570753,0.003888306,0.0007945768,0.0002264091],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9813681,0.0007903022,0.01158635,0.001080473,0.0002844438,0.0002963941,0.000007493077,0.0001067864,0.004479666],"genre_scores_gemma":[0.9971799,0.0006752923,0.001424621,0.0001149656,0.0004066515,0.0001063499,0.00001397599,0.0000294542,0.00004883997],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7398646,"threshold_uncertainty_score":0.9996127,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3101267683933944,"score_gpt":0.5719112199459427,"score_spread":0.2617844515525484,"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."}}