{"id":"W4394769784","doi":"10.1016/j.acra.2024.02.028","title":"Use of Response Permutation to Measure an Imaging Dataset’s Susceptibility to Overfitting by Selected Standard Analysis Pipelines","year":2024,"lang":"en","type":"article","venue":"Academic Radiology","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University; University Health Network","funders":"National Cancer Institute; National Institutes of Health; Queen's University","keywords":"Overfitting; Artificial intelligence; Random forest; Computer science; Support vector machine; Pattern recognition (psychology); Linear discriminant analysis; Machine learning; Logistic regression; Sample size determination; Cross-validation; Bootstrapping (finance); Receiver operating characteristic; Statistics; Data mining; Mathematics; Artificial neural network; Econometrics","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.150533,0.002350795,0.001270895,0.002333006,0.002264772,0.003499894,0.002928174,0.002719532,0.002053251],"category_scores_gemma":[0.355328,0.0008329644,0.002615711,0.002852761,0.004410233,0.002570788,0.00321188,0.003731679,0.0007624661],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001417085,"about_ca_system_score_gemma":0.002881226,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001043858,"about_ca_topic_score_gemma":0.001256023,"domain_scores_codex":[0.8613853,0.1012117,0.009574237,0.01247409,0.01403971,0.001314905],"domain_scores_gemma":[0.6958938,0.1928766,0.02655318,0.06287486,0.02033689,0.001464651],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.007884784,0.003035383,0.3136029,0.003038516,0.006002116,0.001492934,0.004847336,0.06032507,0.09026001,0.03420543,0.01504564,0.4602599],"study_design_scores_gemma":[0.001719077,0.01143909,0.1871956,0.001051266,0.002663917,0.003421019,0.001325971,0.3788946,0.2732991,0.09355512,0.04458869,0.0008466131],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1009591,0.000399612,0.8882399,0.0008536276,0.0003017801,0.003667473,0.000894161,0.002805682,0.001878576],"genre_scores_gemma":[0.5199691,0.000165681,0.463352,0.001041235,0.0001462656,0.01149583,0.001790695,0.001168013,0.0008711202],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.849467,"threshold_uncertainty_score":0.796104,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02666421758784776,"score_gpt":0.3603955308308786,"score_spread":0.3337313132430308,"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."}}