{"id":"W4390512214","doi":"10.1016/j.artmed.2023.102751","title":"Evaluating the clinical utility of artificial intelligence assistance and its explanation on the glioma grading task","year":2024,"lang":"en","type":"article","venue":"Artificial Intelligence in Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia; University of British Columbia Hospital; Simon Fraser University","funders":"Alliance de recherche numérique du Canada; BC Cancer Foundation; Nvidia","keywords":"Grading (engineering); Computer science; Artificial intelligence; Task (project management); Machine learning; Glioma; Medicine","routes":{"ca_aff":true,"ca_fund":true,"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":[0.003402719,0.0006017378,0.0003960283,0.0008457176,0.0001804814,0.0009764336,0.0006296998,0.001548393,0.002574957],"category_scores_gemma":[0.03510493,0.000120263,0.0004300864,0.0004294982,0.0003555942,0.0008508064,0.0005707012,0.000590146,0.0005405418],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002838625,"about_ca_system_score_gemma":0.0007511195,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003322886,"about_ca_topic_score_gemma":0.00290521,"domain_scores_codex":[0.9987957,0.000707981,0.0001169762,0.0001613875,0.000164841,0.00005310921],"domain_scores_gemma":[0.9688922,0.02814165,0.000749648,0.0006686636,0.001071442,0.0004765548],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.03337768,0.003431556,0.1769864,0.00126623,0.0009948683,0.0007921483,0.00115427,0.04708395,0.01045636,0.0009767825,0.01415277,0.709327],"study_design_scores_gemma":[0.003361654,0.01072205,0.231596,0.0004291618,0.00250246,0.001521766,0.0013778,0.7113841,0.0205819,0.006638474,0.00965761,0.0002271499],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.985333,0.001898497,0.007121535,0.00110924,0.0001716801,0.0002145296,0.0009038207,0.0004490093,0.002798649],"genre_scores_gemma":[0.9932771,0.0003586478,0.004412316,0.0001352723,0.0000708234,0.00004659473,0.0008475115,0.00002614859,0.0008254877],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003402719,"threshold_uncertainty_score":0.01799554,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5735416713807948,"score_gpt":0.5688673115818328,"score_spread":0.004674359798962024,"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."}}