{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5016562,0.00145614,0.004164213,0.02610896,0.005393894,0.008767208,0.006382681,0.005105689,0.004288536],"category_scores_gemma":[0.8333657,0.002671226,0.004200634,0.01891211,0.02240271,0.01703166,0.0116964,0.007238288,0.0006775729],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006228419,"about_ca_system_score_gemma":0.009391206,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005092114,"about_ca_topic_score_gemma":0.003442822,"domain_scores_codex":[0.2251874,0.6646104,0.0342638,0.0191274,0.05463159,0.002179432],"domain_scores_gemma":[0.02385804,0.9109249,0.01868659,0.02572123,0.02006514,0.0007440507],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.001367138,0.0008068302,0.3514411,0.009839845,0.01018393,0.0006937623,0.03707882,0.007982107,0.0008413378,0.3913778,0.003890656,0.1844967],"study_design_scores_gemma":[0.0004094513,0.002958141,0.2717995,0.01469419,0.005017526,0.004349203,0.02206278,0.0317482,0.006437837,0.5989473,0.04086218,0.0007137567],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.2635214,0.1388338,0.5097904,0.01081575,0.003099205,0.002358455,0.001002956,0.0002558396,0.0703222],"genre_scores_gemma":[0.9146019,0.006400105,0.07265326,0.001747393,0.001318847,0.001547048,0.0002860646,0.0001450412,0.001300375],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.4983438,"threshold_uncertainty_score":0.6145467,"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."}}