{"id":"W7116657517","doi":"10.1186/s13244-025-02158-4","title":"Skull-stripping induces shortcut learning in MRI-based Alzheimer’s disease classification","year":2025,"lang":"en","type":"article","venue":"Insights into Imaging","topic":"Dementia and Cognitive Impairment Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institutes of Health Research; National Institutes of Health; Genentech; IXICO; H. Lundbeck A/S; Servier; Eisai; Austrian Science Fund; Northern California Institute for Research and Education; Pfizer; Novartis Pharmaceuticals Corporation; University of Southern California; Biogen; Eli Lilly and Company; Bristol-Myers Squibb; BioClinica; U.S. Department of Defense; Alzheimer's Disease Neuroimaging Initiative; Meso Scale Diagnostics; National Institute on Aging; Alzheimer's Association","keywords":"Deep learning; Preprocessor; Interpretation (philosophy); Cluster analysis; Neuroradiology; Disease","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000293272,0.0001734722,0.0002132077,0.0007980441,0.0002085446,0.0001048599,0.0001283982,0.00004308491,0.00009588718],"category_scores_gemma":[0.0002602615,0.0001572974,0.00007963119,0.0006698893,0.0001364723,0.00030731,0.00007255406,0.0004670711,0.00003989735],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001703958,"about_ca_system_score_gemma":0.0005202567,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001165917,"about_ca_topic_score_gemma":0.00002309601,"domain_scores_codex":[0.9983549,0.0001455046,0.0003386575,0.0004320143,0.0003890679,0.0003398423],"domain_scores_gemma":[0.9992255,0.0001211108,0.00006278311,0.0002350064,0.0001926109,0.0001629841],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002255011,0.0003027439,0.8317624,0.0001189098,0.00007483715,0.0002430035,0.0005310206,0.00006191768,0.01935146,0.0008412941,0.000271212,0.1462157],"study_design_scores_gemma":[0.00207923,0.00006053095,0.9007642,0.001105164,0.0002162745,0.000002215785,0.001399135,0.06080087,0.01529932,0.001047938,0.01697222,0.0002528793],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.948153,0.003876579,0.009571423,0.007941118,0.0002100355,0.0007728772,6.381773e-7,0.0001862534,0.02928806],"genre_scores_gemma":[0.9976113,0.00006305193,0.0004205282,0.001175354,0.00006803892,0.0001036753,0.00005902291,0.00001803112,0.0004810419],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1459628,"threshold_uncertainty_score":0.6414401,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03038588687508028,"score_gpt":0.3488645106658836,"score_spread":0.3184786237908033,"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."}}