{"id":"W4283735617","doi":"10.1002/hed.27130","title":"Augmenting inter‐rater concordance of radiologic extranodal extension in <scp>HPV</scp> ‐positive oropharyngeal carcinoma: A multicenter study","year":2022,"lang":"en","type":"article","venue":"Head & Neck","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":25,"is_retracted":false,"has_abstract":true,"ca_institutions":"Princess Margaret Cancer Centre; University of Toronto","funders":"","keywords":"Certainty; Concordance; Kappa; Inter-rater reliability; Medicine; Internal medicine; Radiology; Nuclear medicine; Psychology; Mathematics; 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":[],"consensus_categories":[],"category_scores_codex":[0.00534215,0.0002478942,0.0006340439,0.0002583755,0.0002719223,0.00007851029,0.0007946966,0.00004620022,0.0005509555],"category_scores_gemma":[0.001165892,0.0001901297,0.0001737859,0.0006623677,0.0001321566,0.0002514459,0.0005941489,0.0004337619,0.00005743402],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002278405,"about_ca_system_score_gemma":0.00007118574,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003039372,"about_ca_topic_score_gemma":0.0002078979,"domain_scores_codex":[0.9940076,0.001563546,0.001303567,0.0009212901,0.001713058,0.0004908904],"domain_scores_gemma":[0.9973251,0.001215104,0.0003955734,0.0006788515,0.0002880116,0.00009733843],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002514793,0.001957363,0.9664021,0.00002052472,0.00002586942,0.0002067755,0.009269217,0.0009566118,0.01401216,0.00002123883,0.001680018,0.005196602],"study_design_scores_gemma":[0.003272249,0.001828028,0.9648969,0.00004886675,0.00002410186,0.00001098102,0.02161955,0.004192603,0.001996269,0.0003995054,0.001592705,0.0001182711],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961598,0.0004821387,0.0001982692,0.0002121116,0.0007771559,0.00121711,0.00002694567,0.00002188922,0.0009045841],"genre_scores_gemma":[0.9986115,0.000005991059,0.0003049044,0.0003477735,0.00005611076,0.000121999,0.000006364851,0.00001308625,0.0005323256],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01235033,"threshold_uncertainty_score":0.7753265,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1017680718029171,"score_gpt":0.3508374837631717,"score_spread":0.2490694119602546,"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."}}