{"id":"W4412163675","doi":"10.1158/1557-3265.aimachine-b008","title":"Abstract B008: Fine-tuned large language model accurately estimates progression-free survival from radiology reports in immune checkpoint inhibitor–treated cancers","year":2025,"lang":"en","type":"article","venue":"Clinical Cancer Research","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; Immune checkpoint; Cancer; Oncology; Internal medicine; Radiology; Cancer research; Immunotherapy","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001877672,0.001301671,0.0008175479,0.0008570841,0.0003331943,0.001003671,0.001587424,0.001240788,0.002208797],"category_scores_gemma":[0.005390943,0.0004362753,0.001507901,0.0004209332,0.0003418305,0.0006597677,0.0006810445,0.00152829,0.0008202486],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001623111,"about_ca_system_score_gemma":0.001518047,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03353506,"about_ca_topic_score_gemma":0.03308219,"domain_scores_codex":[0.9994141,0.0001637049,0.00004185603,0.000248941,0.00005933502,0.00007217666],"domain_scores_gemma":[0.9981988,0.00123134,0.0001309395,0.000146758,0.0002087717,0.00008345942],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001310479,0.0004491204,0.0346269,0.0002476499,0.0003787041,0.000411075,0.00009951698,0.8770041,0.003791347,0.000719748,0.01546219,0.06549919],"study_design_scores_gemma":[0.00006447398,0.00008712614,0.002542656,0.00001494994,0.00004655496,0.00004274658,0.00001285564,0.9948601,0.0008577516,0.0007172362,0.0007391179,0.00001444566],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8541307,0.004198374,0.09140448,0.004201869,0.0004769133,0.0003357248,0.02793927,0.01313925,0.004173425],"genre_scores_gemma":[0.946647,0.000364925,0.02548742,0.0006885824,0.0001215665,0.000187135,0.0234482,0.0002052793,0.002849776],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03353506,"threshold_uncertainty_score":0.06667972,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1072499623074493,"score_gpt":0.5259392477263349,"score_spread":0.4186892854188856,"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."}}