{"id":"W4307233259","doi":"10.1007/s00330-022-09165-9","title":"Algorithmic transparency and interpretability measures improve radiologists’ performance in BI-RADS 4 classification","year":2022,"lang":"en","type":"article","venue":"European Radiology","topic":"AI in cancer detection","field":"Computer Science","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Deutschen Konsortium für Translationale Krebsforschung; Technische Universität München; Deutsche Forschungsgemeinschaft","keywords":"Interpretability; Certainty; Neuroticism; Artificial intelligence; Pearson product-moment correlation coefficient; Medicine; Correlation; Machine learning; Neuroradiology; Personality; Transparency (behavior); Big Five personality traits; Radiology; Computer science; Psychology; Mathematics; Statistics; Social psychology; Neurology; Psychiatry","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009573238,0.0004602175,0.0003075736,0.0009840553,0.0002094309,0.001299819,0.0003227511,0.0008124626,0.001115371],"category_scores_gemma":[0.08649892,0.0002552572,0.0006386958,0.0003446074,0.0005387294,0.001069633,0.001058547,0.0004654768,0.0002212577],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003832912,"about_ca_system_score_gemma":0.0002807838,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000587867,"about_ca_topic_score_gemma":0.0005932681,"domain_scores_codex":[0.9915803,0.004566296,0.0008326512,0.0007482656,0.002010431,0.000262047],"domain_scores_gemma":[0.9035513,0.0684387,0.01822094,0.003997297,0.004625672,0.00116606],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003161096,0.0002513486,0.8524463,0.000338672,0.0004647449,0.000181249,0.002516701,0.004172273,0.01458936,0.000245857,0.0006324052,0.1210001],"study_design_scores_gemma":[0.00003724549,0.001139131,0.9759368,0.00006014538,0.0001837485,0.0005713333,0.0006914008,0.01419428,0.005738909,0.0006748389,0.0007201525,0.00005207239],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9928601,0.0004115576,0.004845126,0.0002165757,0.0000147335,0.00002801529,0.00004634592,0.0001053985,0.001472181],"genre_scores_gemma":[0.9976451,0.00004112986,0.002161757,0.00001672714,0.00001207104,0.000005279524,0.00003116674,0.000006496912,0.00008028107],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009573238,"threshold_uncertainty_score":0.05062872,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02440863810332993,"score_gpt":0.2311921117057944,"score_spread":0.2067834736024644,"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."}}