{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0320803,0.001897166,0.004552013,0.01140881,0.0007498386,0.003964426,0.004014184,0.00265462,0.004544856],"category_scores_gemma":[0.1235563,0.001260429,0.007851445,0.008092819,0.001210072,0.004760727,0.002138721,0.002119623,0.001266774],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004039848,"about_ca_system_score_gemma":0.01620164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009679976,"about_ca_topic_score_gemma":0.01456479,"domain_scores_codex":[0.9899588,0.004633849,0.002873837,0.0007912493,0.001566358,0.0001758871],"domain_scores_gemma":[0.8755795,0.1086864,0.00504561,0.001676879,0.008648118,0.0003635452],"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":[0.000325036,0.0001034702,0.001327796,0.4625067,0.003629623,0.0001401821,0.0005815622,0.00301709,0.0004237767,0.002775412,0.006363162,0.5188062],"study_design_scores_gemma":[0.0001938503,0.0008999558,0.003745386,0.8292661,0.02606585,0.0004680576,0.0009157283,0.004289792,0.001621878,0.005594914,0.1267962,0.000142309],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001086003,0.991913,0.003577319,0.001072519,0.0001777483,0.0007058749,0.0004867545,0.00007219396,0.000908589],"genre_scores_gemma":[0.01313315,0.9734899,0.009425438,0.0008313634,0.0001842225,0.001636146,0.000945597,0.00006080334,0.0002933173],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.0320803,"threshold_uncertainty_score":0.1696589,"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."}}