{"id":"W4413770354","doi":"10.1007/s00256-025-05009-x","title":"Evaluating diagnostic consistency: intra and interobserver agreement in rotator cuff tear classification using ultrasound videos","year":2025,"lang":"en","type":"article","venue":"Skeletal Radiology","topic":"Shoulder Injury and Treatment","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"Alberta Innovates; Canadian Institute for Advanced Research","keywords":"Medicine; Rotator cuff; Orthopedic surgery; Radiology; Ultrasound; Kappa; Consistency (knowledge bases); Surgery; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.0003481361,0.0001519485,0.0003318062,0.0001564817,0.00006949297,0.00001194173,0.00005002294,0.0001184313,0.00008958582],"category_scores_gemma":[0.001310447,0.0001321913,0.0000530814,0.0001684368,0.0001897054,0.00005259047,0.00003383272,0.000195198,0.00001182678],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002343407,"about_ca_system_score_gemma":0.000146473,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008056795,"about_ca_topic_score_gemma":0.00005378426,"domain_scores_codex":[0.9987584,0.0001954657,0.000361728,0.0003528691,0.00009380764,0.0002377503],"domain_scores_gemma":[0.9987189,0.0008749932,0.0000728326,0.0002224799,0.00004762656,0.00006315238],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002700089,0.0004212791,0.8567592,0.0002440608,0.0004189832,0.0001298041,0.0006400326,0.00005182624,0.09430018,0.006682083,0.0002827482,0.03979979],"study_design_scores_gemma":[0.003381254,0.0008749409,0.9833999,0.0002953217,0.0005218977,0.0003380007,0.0005477108,0.002437213,0.004765975,0.002895056,0.0003691681,0.0001735173],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9959236,0.00148101,0.0002670309,0.0006024634,0.000265606,0.0006524471,0.000004610319,0.00002315222,0.0007800549],"genre_scores_gemma":[0.99731,0.0001187837,0.00176358,0.000499145,0.00005530018,0.00007747502,0.00003155895,0.000009017126,0.0001351262],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1266407,"threshold_uncertainty_score":0.5390604,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06258939021094627,"score_gpt":0.3830206514345004,"score_spread":0.3204312612235541,"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."}}