{"id":"W4413050320","doi":"10.1371/journal.pdig.0000980","title":"Development and evaluation of large-language models (LLMs) for oncology: A scoping review","year":2025,"lang":"en","type":"review","venue":"PLOS Digital Health","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hamilton Health Sciences; Population Health Research Institute; McMaster University","funders":"","keywords":"Oncology; Medicine; Internal medicine","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":[0.002327327,0.0002324183,0.002268648,0.0001746774,0.00006954678,0.00001852409,0.00008878861,0.0001175377,0.000008348131],"category_scores_gemma":[0.001375028,0.0001808219,0.0001764702,0.0001833494,0.0000337307,0.00007923601,0.00007795908,0.0003143351,0.000001843993],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004196725,"about_ca_system_score_gemma":0.008054973,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000002927025,"about_ca_topic_score_gemma":0.000003240554,"domain_scores_codex":[0.9977306,0.0001282141,0.001034284,0.00035339,0.0004557548,0.0002977168],"domain_scores_gemma":[0.9984687,0.0003993329,0.0005473725,0.0002204476,0.0001706502,0.000193508],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000002564291,0.00008092241,4.069468e-7,0.4070586,0.00006420818,0.000001033517,0.000110746,2.302796e-7,1.088979e-8,0.00005556208,0.0001977847,0.5924279],"study_design_scores_gemma":[0.0006614153,0.0001137615,2.769408e-7,0.7903118,0.0008984203,0.00002450901,0.00002985273,0.002046496,2.100097e-7,0.00004024471,0.2057547,0.0001183889],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001424869,0.9906371,0.0009558426,0.0004761119,0.00006624638,0.005842891,0.00005568941,0.00002946778,0.001922405],"genre_scores_gemma":[0.00001380326,0.9927994,0.004712921,0.001113904,0.00006078724,0.000483732,0.0006586919,0.00003085539,0.0001258884],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.5923095,"threshold_uncertainty_score":0.9975684,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1294335098202249,"score_gpt":0.4975380856326087,"score_spread":0.3681045758123838,"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."}}