{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.07789108,0.0009861107,0.001424673,0.006620992,0.001583969,0.00349768,0.002004626,0.002249371,0.001242811],"category_scores_gemma":[0.1750549,0.001135399,0.002328508,0.00259279,0.0021558,0.004095374,0.00362616,0.0014259,0.0006100461],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00123349,"about_ca_system_score_gemma":0.00119921,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002360649,"about_ca_topic_score_gemma":0.005212077,"domain_scores_codex":[0.9220652,0.03858946,0.01661332,0.007492147,0.01320326,0.002036636],"domain_scores_gemma":[0.7835771,0.1355931,0.0206494,0.01604662,0.0421027,0.002031045],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002586257,0.0002701264,0.9345279,0.0004979706,0.002521607,0.000333196,0.005743513,0.001853745,0.005849699,0.0009192171,0.001226217,0.04367048],"study_design_scores_gemma":[0.0002171188,0.001253437,0.9219785,0.0004015391,0.001976202,0.002343185,0.006658105,0.04720599,0.01078379,0.00208583,0.004769623,0.0003266104],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9560913,0.002929356,0.03248439,0.000246489,0.0003710832,0.0005822626,0.0006129067,0.000323053,0.006358951],"genre_scores_gemma":[0.9823741,0.000276171,0.01589225,0.00008395375,0.00009472812,0.0002680871,0.0004289245,0.0001253459,0.0004565344],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07789108,"threshold_uncertainty_score":0.4119323,"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."}}