{"id":"W4386570124","doi":"10.1016/j.ejrad.2023.111091","title":"Inter-reader variability and reproducibility of the PI-QUAL score in a multicentre setting","year":2023,"lang":"en","type":"article","venue":"European Journal of Radiology","topic":"Radiology practices and education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":false,"ca_institutions":"Princess Margaret Cancer Centre; Women's College Hospital; University of Toronto; University Health Network; Mount Sinai Hospital","funders":"","keywords":"Medicine; Reproducibility; Technician; Medical physics; Quality Score; Second opinion; Radiology; Nuclear medicine; Statistics; Pathology","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":[0.01035206,0.0000724793,0.0002793777,0.0000962864,0.00003189948,0.00000363012,0.000123693,0.00003422336,0.00002349215],"category_scores_gemma":[0.009078542,0.00004543709,0.00006681912,0.0001976692,0.0002129778,0.0000692474,0.0000640811,0.0004336961,0.000004417728],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003862348,"about_ca_system_score_gemma":0.0000896067,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00000682385,"about_ca_topic_score_gemma":0.000002458331,"domain_scores_codex":[0.996856,0.002074522,0.000574059,0.0002759882,0.00006972084,0.0001496668],"domain_scores_gemma":[0.9983566,0.0005366058,0.0004510069,0.0004940744,0.000104203,0.00005744633],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003246982,0.00008816255,0.9785729,0.0000632955,0.00004988766,0.00008294338,0.004730808,0.00004807894,0.004004803,0.00002962535,0.002438312,0.009566494],"study_design_scores_gemma":[0.0008752007,0.0002583121,0.9939476,0.00008962649,0.00005244351,0.001579376,0.001182258,0.0001995725,0.0001469107,0.00007772131,0.001554683,0.00003629234],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9884349,0.0001337538,0.00004293771,0.009986545,0.0007849134,0.0001320548,0.000001144257,0.000006481996,0.000477316],"genre_scores_gemma":[0.9988738,0.00006410922,0.0004386067,0.0002421949,0.0002739779,3.589351e-7,0.000001396091,0.000008740774,0.00009675645],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01537471,"threshold_uncertainty_score":0.9992684,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04447527087974113,"score_gpt":0.3187611167516928,"score_spread":0.2742858458719517,"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."}}