{"id":"W4312032523","doi":"10.1101/2022.12.07.22282726","title":"Evaluating the Clinical Utility of Artificial Intelligence Assistance and its Explanation on Glioma Grading Task","year":2022,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; Simon Fraser University","funders":"Alliance de recherche numérique du Canada; BC Cancer Foundation; Nvidia","keywords":"Grading (engineering); Artificial intelligence; Computer science; Glioma; Machine learning; Task (project management); Medical physics; Natural language processing; 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":[0.003076077,0.0009351145,0.0004694159,0.0006831529,0.0002068066,0.0008661225,0.0008894222,0.001620395,0.002391652],"category_scores_gemma":[0.02134796,0.0001857167,0.000618517,0.0003230784,0.0004013781,0.0008333622,0.0006537797,0.001222605,0.0006314559],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005708987,"about_ca_system_score_gemma":0.0007328486,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003906317,"about_ca_topic_score_gemma":0.003524671,"domain_scores_codex":[0.9987203,0.000602183,0.0001041026,0.0003158264,0.0001832868,0.00007441326],"domain_scores_gemma":[0.9871008,0.01055832,0.0006499818,0.0005977754,0.0006706151,0.0004224919],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.009973372,0.002607259,0.2082889,0.00184064,0.0009215305,0.0006230633,0.001017294,0.1258703,0.01255125,0.0009089574,0.02888482,0.6065126],"study_design_scores_gemma":[0.0007833546,0.003676143,0.08230078,0.0002489359,0.0006411272,0.0004850604,0.000446849,0.884627,0.0174553,0.002942637,0.006243699,0.0001490405],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9681158,0.002790354,0.01949657,0.001768824,0.0002346119,0.0003031974,0.001566882,0.002803274,0.002920529],"genre_scores_gemma":[0.9864058,0.0002570854,0.01021815,0.0002187909,0.00004893597,0.0000835695,0.001977899,0.00003818465,0.0007516257],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003906317,"threshold_uncertainty_score":0.01626801,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5535098137900805,"score_gpt":0.5588842013511088,"score_spread":0.005374387561028371,"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."}}