{"id":"W4410230129","doi":"10.1371/journal.pone.0323262","title":"Reliability and agreement during the Rapid Entire Body Assessment: Comparing rater expertise and artificial intelligence","year":2025,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Saskatoon Medical Imaging; Saskatoon City Hospital; University of Saskatchewan","funders":"","keywords":"Trunk; Reliability (semiconductor); Inter-rater reliability; Suite; Intra-rater reliability; Physical therapy; Medicine; Computer science; Physical medicine and rehabilitation; Psychology; Confidence interval; Biology; 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":[],"consensus_categories":[],"category_scores_codex":[0.06149569,0.0005042439,0.0007531124,0.002185851,0.0006703319,0.001486442,0.0007666317,0.0006028067,0.0006974097],"category_scores_gemma":[0.1466811,0.0003994133,0.001028327,0.0008762171,0.001179623,0.001520742,0.001826274,0.0005914273,0.0004131853],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005140234,"about_ca_system_score_gemma":0.0005200444,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001849518,"about_ca_topic_score_gemma":0.002942883,"domain_scores_codex":[0.9331399,0.04118583,0.006374019,0.005268219,0.01311009,0.0009220655],"domain_scores_gemma":[0.8029153,0.1335295,0.01655785,0.01140951,0.03446663,0.001121233],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002414599,0.0002519804,0.7982063,0.0009257146,0.00231901,0.0003239253,0.03300694,0.003433366,0.0126314,0.001084242,0.001460413,0.143942],"study_design_scores_gemma":[0.0001426063,0.002251833,0.9425256,0.0005455613,0.0007314118,0.001566563,0.009248444,0.02419669,0.01118464,0.002184333,0.00515609,0.0002662342],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9503161,0.001191274,0.04154847,0.0001484536,0.0001750037,0.0003438956,0.0002287894,0.0001498637,0.005898199],"genre_scores_gemma":[0.9898622,0.0002232855,0.008920913,0.00005486227,0.00005193894,0.0001756061,0.0001739749,0.0000585118,0.0004786989],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06149569,"threshold_uncertainty_score":0.3252242,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03429571894064445,"score_gpt":0.2969069289628076,"score_spread":0.2626112100221632,"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."}}