{"id":"W4412124420","doi":"10.2196/71176","title":"Language Models for Multilabel Document Classification of Surgical Concepts in Exploratory Laparotomy Operative Notes: Algorithm Development Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Digital Imaging in Medicine","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Institute of General Medical Sciences","keywords":"Exploratory laparotomy; Computer science; Natural language processing; Development (topology); Artificial intelligence; Information retrieval; Algorithm; Medicine; Surgery; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"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":[0.004940151,0.001123012,0.0009941809,0.001729899,0.0005706681,0.001692145,0.001700828,0.00150481,0.002201132],"category_scores_gemma":[0.01094529,0.0003390384,0.001146428,0.001279142,0.0003452767,0.0017189,0.001130997,0.002291746,0.001408448],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001522223,"about_ca_system_score_gemma":0.001704338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007709737,"about_ca_topic_score_gemma":0.007233866,"domain_scores_codex":[0.9984383,0.0005871254,0.0001763223,0.0004440036,0.0002320553,0.0001221635],"domain_scores_gemma":[0.9912141,0.006618055,0.0003320412,0.000283306,0.001407983,0.0001445602],"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":[0.00059966,0.0006030407,0.007177467,0.0002866361,0.0002368208,0.0001582346,0.0002651238,0.168863,0.004369298,0.002312331,0.005940024,0.8091884],"study_design_scores_gemma":[0.00002870592,0.00008488453,0.0006535541,0.00002504508,0.00003660642,0.00006331691,0.00007110485,0.9952044,0.001818086,0.001377986,0.0006257126,0.00001058795],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1271736,0.00228869,0.8623066,0.0008147349,0.0001503065,0.0004724225,0.0009602024,0.004626483,0.001206941],"genre_scores_gemma":[0.3731866,0.0008025122,0.6181041,0.0003857386,0.0001318314,0.0007794451,0.003699851,0.0002236517,0.002686195],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.007709737,"threshold_uncertainty_score":0.02612633,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03512460637774333,"score_gpt":0.4007315009985608,"score_spread":0.3656068946208174,"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."}}