{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003789241,0.0007948792,0.0005457495,0.0008743065,0.0003918832,0.0009560069,0.0008382538,0.0008597611,0.001100751],"category_scores_gemma":[0.02183988,0.000350616,0.0006916994,0.0004851461,0.001161896,0.0008365482,0.001315497,0.0009006063,0.0003266515],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005550298,"about_ca_system_score_gemma":0.0007485072,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001475344,"about_ca_topic_score_gemma":0.001834372,"domain_scores_codex":[0.999015,0.0003525696,0.00007495344,0.0002483416,0.0002441244,0.00006507058],"domain_scores_gemma":[0.993646,0.003660641,0.001057142,0.0006936624,0.0007594264,0.0001830544],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002121366,0.0003005865,0.05769706,0.0008356246,0.0007071455,0.0009827701,0.0005900362,0.2136096,0.09214335,0.004745785,0.004776712,0.6214899],"study_design_scores_gemma":[0.00006062329,0.0007424992,0.04098515,0.0001176906,0.000241362,0.0006295129,0.0001598871,0.8731863,0.0668284,0.0149948,0.002002193,0.00005168707],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7433677,0.00174291,0.2511952,0.0005637781,0.0001496297,0.0001797417,0.0002734421,0.0009232158,0.001604436],"genre_scores_gemma":[0.960129,0.0004359762,0.0379874,0.0001101425,0.00006601581,0.00005124463,0.0003882278,0.0001130473,0.0007188533],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003789241,"threshold_uncertainty_score":0.02003968,"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."}}