{"id":"W3199377622","doi":"10.1002/pmrj.12705","title":"Reliability and validity of subjective radiologist reporting of temporal changes in lumbar spine <scp>MRI</scp> findings","year":2021,"lang":"en","type":"article","venue":"PM&R","topic":"Radiology practices and education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Health and Medical Research Council; National Institutes of Health","keywords":"Medicine; Magnetic resonance imaging; Inter-rater reliability; Lumbar; Low back pain; Radiology; Kappa; Reliability (semiconductor); False positive paradox; Physical therapy; Physical medicine and rehabilitation; Psychology; Pathology; Rating scale","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":[0.001785866,0.00007843964,0.0004325954,0.00006342088,0.00002570555,0.000003284253,0.00003190648,0.0001121122,0.00001512327],"category_scores_gemma":[0.01549561,0.00006986407,0.00004238324,0.0002040283,0.0001582981,0.00007002201,0.00003308413,0.0001824181,5.427539e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005079932,"about_ca_system_score_gemma":0.0001415103,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003633704,"about_ca_topic_score_gemma":0.0001616425,"domain_scores_codex":[0.9988125,0.0001282746,0.0005741593,0.000253969,0.00008967405,0.0001414348],"domain_scores_gemma":[0.9985499,0.0002564286,0.0007321796,0.0002314236,0.0001819001,0.00004810297],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001504502,0.0001303194,0.9602118,0.0002226264,0.00001980987,0.00003237946,0.001287761,0.000004318772,0.03740082,0.0000303253,0.0003044497,0.0003403596],"study_design_scores_gemma":[0.0003636788,0.0002259024,0.9582358,0.00007507556,0.00004759614,0.0001651064,0.001721535,0.00005967341,0.03819215,0.0002059866,0.0006822081,0.00002530941],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9968389,0.0005157528,0.00002394491,0.001417519,0.0001695876,0.0001663098,0.000003721192,0.00001000293,0.0008542379],"genre_scores_gemma":[0.9974102,0.0004111033,0.001697718,0.00006083706,0.00009579384,0.000008648383,0.00002840383,0.000006046179,0.0002812137],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01370974,"threshold_uncertainty_score":0.9927973,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05999772335550631,"score_gpt":0.3460465078143599,"score_spread":0.2860487844588536,"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."}}