{"id":"W4408776600","doi":"10.1016/j.labinv.2024.102237","title":"15 Identifying the Optimal Architecture for Using Local Large Language Models (LLM) In Autopsy Pathology Reports","year":2025,"lang":"en","type":"article","venue":"Laboratory Investigation","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Institute for Research in Immunology and Cancer; Centre Hospitalier de l’Université de Montréal; Université de Montréal","funders":"","keywords":"Autopsy; Pathology; Architecture; Computer science; Medicine; Natural language processing; Geography; Archaeology","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":[0.001291407,0.00013703,0.0001642725,0.0001961665,0.0002402707,0.0001089669,0.0003658755,0.0001382773,0.000002065286],"category_scores_gemma":[0.0003528159,0.0001182252,0.00003531347,0.0009103681,0.0000890486,0.0003383205,0.0001668497,0.0004091682,0.00000109033],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001203085,"about_ca_system_score_gemma":0.0004586367,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000106745,"about_ca_topic_score_gemma":0.0002246575,"domain_scores_codex":[0.9982606,0.000412662,0.0003864302,0.0004420042,0.0001822161,0.0003161076],"domain_scores_gemma":[0.9988405,0.0001702619,0.0001925255,0.0005904053,0.0001495652,0.00005670763],"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":[0.00001605849,0.00005695067,0.02206447,0.0005038532,0.00002288643,0.0003244437,0.02804993,0.6459811,0.008435333,0.2847763,0.0004279927,0.009340656],"study_design_scores_gemma":[0.0002279213,0.00002521196,0.004130583,0.0001232539,0.000007788753,0.00002817549,0.000453694,0.9574111,0.002380793,0.0345984,0.0004705063,0.0001425853],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3761162,0.0003540082,0.6212059,0.001532265,0.0003556224,0.0002973258,0.000006183698,0.0001001217,0.00003234499],"genre_scores_gemma":[0.9121192,0.000001002033,0.08518287,0.002503571,0.00006197135,0.00006886942,0.00001750986,0.00001222459,0.00003276625],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.536023,"threshold_uncertainty_score":0.4821085,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02274256757964648,"score_gpt":0.3179683781641615,"score_spread":0.295225810584515,"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."}}