{"id":"W4399781895","doi":"10.1148/radiol.233117","title":"Large Language Models for Automated Synoptic Reports and Resectability Categorization in Pancreatic Cancer","year":2024,"lang":"en","type":"article","venue":"Radiology","topic":"Radiology practices and education","field":"Medicine","cited_by":75,"is_retracted":false,"has_abstract":true,"ca_institutions":"Princess Margaret Cancer Centre; Toronto General Hospital; University of Toronto; University Health Network","funders":"","keywords":"Medicine; Categorization; Cancer; Radiology; Pancreatic cancer; Natural language processing; Medical physics; Artificial intelligence; Oncology; General surgery; 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":[0.004602013,0.001230333,0.0005754445,0.001818067,0.0006862345,0.002041202,0.001745553,0.001296977,0.003468964],"category_scores_gemma":[0.01468733,0.0005243586,0.001824841,0.000862854,0.0004037684,0.001978908,0.001451945,0.002339766,0.002718606],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001865152,"about_ca_system_score_gemma":0.002564013,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01749393,"about_ca_topic_score_gemma":0.02719801,"domain_scores_codex":[0.9981297,0.0008450137,0.0002090287,0.0004564036,0.0002314625,0.0001284816],"domain_scores_gemma":[0.987125,0.009986174,0.0005811974,0.0005970614,0.001420043,0.0002904494],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002812516,0.001151331,0.03252644,0.001214762,0.0007127462,0.001877901,0.001667218,0.2688603,0.01376863,0.009236579,0.07897225,0.5871993],"study_design_scores_gemma":[0.00006821536,0.0000909517,0.001631763,0.00005873414,0.00008547764,0.0001568516,0.0001403404,0.9863567,0.002330316,0.004963313,0.00408181,0.00003548098],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3327689,0.003589934,0.5809466,0.005934276,0.0009395874,0.001152666,0.0268921,0.04242684,0.005348966],"genre_scores_gemma":[0.6430073,0.0008083016,0.3187972,0.0009087858,0.0004255163,0.0008629694,0.02989206,0.0009002408,0.004397576],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01749393,"threshold_uncertainty_score":0.0347842,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02171129996342566,"score_gpt":0.3599718246146089,"score_spread":0.3382605246511833,"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."}}