{"id":"W4417346112","doi":"10.64898/2025.12.11.25342109","title":"Deceptive Bias Measurement in Deep Learning: Assessing Shortcut Reliance in TCGA Cancer Models","year":2025,"lang":"","type":"article","venue":"medRxiv","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Brock University; Ontario Tech University","funders":"","keywords":"Metric (unit); Exploit; Limit (mathematics); Measure (data warehouse); Key (lock); Deep learning","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":[0.0151597,0.001243371,0.001014645,0.001270989,0.000681136,0.001667062,0.001764902,0.0016619,0.001098444],"category_scores_gemma":[0.07925854,0.0004826196,0.0007307425,0.0006685351,0.002411232,0.001995767,0.002933607,0.003345377,0.0002147622],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001764357,"about_ca_system_score_gemma":0.001191253,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004262675,"about_ca_topic_score_gemma":0.003801692,"domain_scores_codex":[0.9949992,0.002502613,0.0003351505,0.000753237,0.001061406,0.0003484802],"domain_scores_gemma":[0.9126938,0.06494954,0.007152817,0.007842374,0.004852455,0.002508965],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001298084,0.0002414012,0.07063255,0.0002583714,0.0004139828,0.0002992546,0.0004381255,0.8721839,0.00421516,0.008264035,0.002444501,0.0393106],"study_design_scores_gemma":[0.00001849656,0.0002174535,0.004332246,0.00004429962,0.00003615892,0.0000693263,0.00004830538,0.9833515,0.002509879,0.009102795,0.0002467865,0.00002279175],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8102545,0.001321482,0.1835946,0.001447056,0.00008345693,0.00009091604,0.00039789,0.0005908774,0.002219334],"genre_scores_gemma":[0.9921426,0.00008822336,0.006892113,0.0001764877,0.00002058622,0.00003459806,0.0002500016,0.0000489557,0.0003464496],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0151597,"threshold_uncertainty_score":0.08017319,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1439787315026596,"score_gpt":0.3484576135299461,"score_spread":0.2044788820272865,"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."}}