{"id":"W4403307412","doi":"10.2196/58666","title":"Facilitating Trust Calibration in Artificial Intelligence–Driven Diagnostic Decision Support Systems for Determining Physicians’ Diagnostic Accuracy: Quasi-Experimental Study","year":2024,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Japan Society for the Promotion of Science","keywords":"Preprint; Medical diagnosis; Calibration; Diagnostic accuracy; Computer science; Artificial intelligence; Medical physics; Telepathology; Data mining; Machine learning; Medicine; Statistics; World Wide Web; Pathology; Mathematics; Radiology; Telemedicine","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":[0.028256,0.001251821,0.001255037,0.0007808149,0.002052697,0.001417775,0.001589851,0.001638469,0.004807368],"category_scores_gemma":[0.04858728,0.001918718,0.001219392,0.0004226515,0.003443224,0.002126292,0.001571195,0.002173596,0.0005752189],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001662357,"about_ca_system_score_gemma":0.003350046,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009558229,"about_ca_topic_score_gemma":0.001279782,"domain_scores_codex":[0.9808179,0.01277066,0.001595771,0.001737618,0.00169271,0.001385389],"domain_scores_gemma":[0.9283785,0.04252834,0.01159075,0.007636223,0.004931993,0.004934228],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"nonrandomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.05218858,0.823661,0.04903951,0.001346898,0.0003910091,0.0003017081,0.02249095,0.001499366,0.009015708,0.001380548,0.0009237046,0.03776111],"study_design_scores_gemma":[0.06085913,0.8206191,0.08862341,0.0002759491,0.0004903284,0.000179901,0.007191936,0.008633878,0.007573776,0.001743914,0.003573307,0.000235364],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9947299,0.0000278586,0.001102548,0.00007184447,0.00004289698,0.003634291,0.00003368134,0.0000130398,0.0003438597],"genre_scores_gemma":[0.9722428,0.00008314548,0.01021114,0.0002544303,0.000108064,0.01636894,0.00008693767,0.000009076114,0.0006355009],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.028256,"threshold_uncertainty_score":0.1494338,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1264424454375611,"score_gpt":0.4933895653315652,"score_spread":0.3669471198940041,"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."}}