{"id":"W4412163750","doi":"10.1158/1557-3265.aimachine-b019","title":"Abstract B019: Clinician-AI evaluation of prognostic information extraction in head and neck cancer using an on-premises LLM","year":2025,"lang":"en","type":"article","venue":"Clinical Cancer Research","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Jewish General Hospital; University of Calgary; McGill University Health Centre; McGill University","funders":"","keywords":"Premises; Head and neck cancer; Medicine; Head and neck; Cancer; Oncology; Internal medicine; Surgery","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0034714,0.0005809806,0.00035519,0.001006083,0.0004532117,0.001060101,0.001273658,0.0009766504,0.006524517],"category_scores_gemma":[0.01405397,0.0002156971,0.0006498001,0.0006870702,0.0004451469,0.0009353613,0.001849415,0.0007448661,0.002639065],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00165457,"about_ca_system_score_gemma":0.001684228,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0116053,"about_ca_topic_score_gemma":0.01610047,"domain_scores_codex":[0.9980527,0.0007425287,0.0002298459,0.0005415305,0.0003143183,0.0001190148],"domain_scores_gemma":[0.9941596,0.003156738,0.00031226,0.001163556,0.0008091239,0.0003987354],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.009472227,0.003569046,0.2268327,0.002296164,0.0006250925,0.001788069,0.001446221,0.08181553,0.01448374,0.003517546,0.1308844,0.5232693],"study_design_scores_gemma":[0.001228662,0.002464747,0.1048693,0.0003346094,0.0001988025,0.001364276,0.001662546,0.7986588,0.02697193,0.004873371,0.05724282,0.0001301919],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.8699454,0.0009599318,0.04900728,0.002500197,0.0004422039,0.001622169,0.04929546,0.01654738,0.00968003],"genre_scores_gemma":[0.8504182,0.0001954584,0.08837435,0.0004892665,0.00009020456,0.0006168209,0.05597216,0.0001939187,0.003649611],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.0116053,"threshold_uncertainty_score":0.02307546,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2956370655473759,"score_gpt":0.6309681627015147,"score_spread":0.3353310971541388,"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."}}