{"id":"W2500917129","doi":"10.1016/j.diii.2016.05.014","title":"Agreement studies in radiology research","year":2016,"lang":"en","type":"article","venue":"Diagnostic and Interventional Imaging","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":48,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta Hospital; Université de Montréal; Hôpital Notre-Dame; Health Sciences Centre; Centre Hospitalier de l’Université de Montréal","funders":"","keywords":"Medicine; Confidence interval; Medical physics; Reliability (semiconductor); Sample size determination; Nuclear medicine; Statistics; Internal medicine","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01137222,0.00008949568,0.0001970586,0.0003396524,0.0001198375,0.0000683927,0.0002986299,0.00002023692,0.0008703622],"category_scores_gemma":[0.03050197,0.00004924597,0.00008179222,0.0002554522,0.0005667501,0.0002633537,0.0002844772,0.000107922,0.0003159951],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001185455,"about_ca_system_score_gemma":0.00002200383,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002622939,"about_ca_topic_score_gemma":0.00004732952,"domain_scores_codex":[0.9969463,0.0006546952,0.0006305856,0.0004392775,0.001036883,0.0002922488],"domain_scores_gemma":[0.9868995,0.01222662,0.00007974542,0.0002356279,0.0004949858,0.00006354898],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002214012,0.0002192602,0.7568604,0.00001852466,0.00002932751,0.00003881653,0.0003844635,0.000005895562,0.0004743188,0.01737137,0.03482472,0.1897507],"study_design_scores_gemma":[0.00105732,0.0001834781,0.5171729,0.001054122,0.000006353092,0.0000226497,0.003078101,0.0001218991,0.0002873397,0.4556624,0.0211958,0.000157575],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9513464,0.01385302,0.004425569,0.02610916,0.0009188117,0.0003620393,0.00001190185,0.00001384226,0.002959252],"genre_scores_gemma":[0.997516,0.0006429974,0.0001585512,0.0001573804,0.00008025744,0.00005795064,7.861142e-7,0.000003096834,0.001382982],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.438291,"threshold_uncertainty_score":0.9776645,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.512024444147994,"score_gpt":0.5438569514778593,"score_spread":0.03183250732986531,"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."}}