{"id":"W4402297886","doi":"10.1016/j.xcrm.2024.101713","title":"Reconstruction of patient-specific confounders in AI-based radiologic image interpretation using generative pretraining","year":2024,"lang":"en","type":"article","venue":"Cell Reports Medicine","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Leeds Biomedical Research Centre; HORIZON EUROPE Framework Programme; National Cancer Institute; National Institutes of Health; Gemeinsame Bundesausschuss; Bundesministerium für Gesundheit; Deutscher Akademischer Austauschdienst; Deutsche Krebshilfe; European Commission; Bundesministerium für Bildung und Forschung; National Institute for Health and Care Research","keywords":"Confounding; Robustness (evolution); Reliability (semiconductor); Artificial intelligence; Medicine; Computer science; Machine learning; 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":[],"consensus_categories":[],"category_scores_codex":[0.000768802,0.0001336971,0.0002643191,0.0003801094,0.00004629985,0.00003523965,0.000117598,0.00007674064,0.00006123272],"category_scores_gemma":[0.0002531738,0.000113834,0.00004108933,0.0005186271,0.0001795826,0.0002589185,0.00003210205,0.0002857728,0.000001134168],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001924227,"about_ca_system_score_gemma":0.0002277812,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001273913,"about_ca_topic_score_gemma":0.000006130822,"domain_scores_codex":[0.9981216,0.0002166286,0.0007416839,0.0004695824,0.0002649197,0.0001855949],"domain_scores_gemma":[0.9989092,0.0002309709,0.000328187,0.000339901,0.0001304193,0.00006129801],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00008204875,0.00009307845,0.04592933,0.00186102,0.00004990106,0.004392033,0.06473304,0.075406,0.1393562,0.003337238,0.0007774979,0.6639826],"study_design_scores_gemma":[0.0001838714,0.0003002589,0.0006528856,0.0009407955,0.000007078135,0.0004447999,0.0006758794,0.9903264,0.004495336,0.001632021,0.0002124576,0.0001282335],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.342685,0.001615377,0.6501347,0.0007583501,0.002366478,0.0002798448,3.742576e-7,0.0001116742,0.002048218],"genre_scores_gemma":[0.9627512,0.00001854702,0.0369157,0.0001863105,0.00008523824,0.000007678693,0.000009941334,0.00001067892,0.00001467845],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9149204,"threshold_uncertainty_score":0.4642018,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02527191272522262,"score_gpt":0.3063805279896156,"score_spread":0.281108615264393,"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."}}