{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06961718,0.0007034698,0.001210602,0.004718947,0.0007516447,0.002055426,0.001166072,0.001313232,0.0008591929],"category_scores_gemma":[0.1967755,0.0004068722,0.001097585,0.002635967,0.0009987992,0.00144682,0.001783611,0.001277454,0.0005748015],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001317243,"about_ca_system_score_gemma":0.0006018282,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001821013,"about_ca_topic_score_gemma":0.003374262,"domain_scores_codex":[0.9323927,0.03836942,0.008470323,0.005823762,0.01402958,0.0009143214],"domain_scores_gemma":[0.7303195,0.2201478,0.01368023,0.01500303,0.01975256,0.00109691],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003945369,0.0001675599,0.6301914,0.001037065,0.0016842,0.0003722896,0.002661825,0.00807669,0.01020774,0.001938152,0.007893455,0.3318242],"study_design_scores_gemma":[0.0001196736,0.001059182,0.9227692,0.0003710815,0.0008608502,0.004054428,0.001034309,0.03767777,0.01329178,0.004327192,0.01419516,0.0002392367],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8260217,0.0307873,0.1198453,0.001221904,0.001224491,0.0004830552,0.00194059,0.00119178,0.01728387],"genre_scores_gemma":[0.9780027,0.0009673689,0.01801559,0.0003059517,0.0002978414,0.000184701,0.001074995,0.0003528382,0.0007980055],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06961718,"threshold_uncertainty_score":0.3681753,"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."}}