{"id":"W2808749774","doi":"10.1016/j.acra.2018.05.017","title":"Interobserver Variation in Response Evaluation Criteria in Solid Tumors 1.1","year":2018,"lang":"en","type":"article","venue":"Academic Radiology","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":false,"ca_institutions":"Tellabs (Canada)","funders":"","keywords":"Medicine; Kappa; Response Evaluation Criteria in Solid Tumors; Confidence interval; Radiological weapon; Nuclear medicine; Cohen's kappa; Waiver; Clinical trial; Radiology; Internal medicine; Statistics","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":[0.004977908,0.0001160447,0.0003362827,0.0008150974,0.00002645979,0.000004888966,0.0001356667,0.0002760629,0.0002451544],"category_scores_gemma":[0.004976932,0.0001070994,0.00003239996,0.0005390923,0.000177041,0.00009830589,0.00004139074,0.0009592445,0.00005131963],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002439708,"about_ca_system_score_gemma":0.0001467643,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000155641,"about_ca_topic_score_gemma":0.00001753165,"domain_scores_codex":[0.9978033,0.0008594703,0.0005224564,0.0003264578,0.0001733865,0.0003149588],"domain_scores_gemma":[0.9992715,0.0002777952,0.0001079392,0.0001919349,0.00006967037,0.00008113922],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.004870573,0.0001122332,0.3249478,0.00005006151,0.0000496173,0.0001575219,0.01036771,0.0002914573,0.6236305,0.0007998857,0.01028714,0.02443552],"study_design_scores_gemma":[0.00248436,0.0003066832,0.864633,0.0001821841,0.00002168284,0.0003046304,0.00006480546,0.1277909,0.0003517068,0.001580166,0.002177656,0.0001022746],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.988266,0.0002094678,0.0004823374,0.009611424,0.0004946991,0.0003129059,5.952011e-7,0.00002948328,0.0005930773],"genre_scores_gemma":[0.9955203,0.00009502991,0.0008815171,0.002835979,0.0004814237,0.00003567888,0.00001567991,0.00001786763,0.0001165507],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6232788,"threshold_uncertainty_score":0.5958212,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03315993301833307,"score_gpt":0.3957272107499973,"score_spread":0.3625672777316643,"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."}}