{"id":"W4416329793","doi":"10.1088/3049-477x/ae209d","title":"Integrating explainability and bias detection in binary medical image classification: a systematic review","year":2025,"lang":"en","type":"article","venue":"Machine Learning Health","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Interpretability; Debiasing; Context (archaeology); Counterfactual thinking; Binary classification; Binary number; Feature (linguistics); Adversarial system","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03870081,0.001465559,0.004466225,0.007934029,0.0006104259,0.00449746,0.002500162,0.002329091,0.004223821],"category_scores_gemma":[0.2491221,0.0009934854,0.007481218,0.005864277,0.002306458,0.004020864,0.002145072,0.002064705,0.0004421978],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002940679,"about_ca_system_score_gemma":0.01086993,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004679376,"about_ca_topic_score_gemma":0.009683912,"domain_scores_codex":[0.9676657,0.01577926,0.00829963,0.002452199,0.005502149,0.0003010091],"domain_scores_gemma":[0.6544943,0.316617,0.01675475,0.004287597,0.007391727,0.0004547147],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0002637551,0.0000314661,0.003753218,0.6500633,0.01151067,0.00007424719,0.0004751125,0.0007341065,0.0001466745,0.002778087,0.004447925,0.3257214],"study_design_scores_gemma":[0.0002100759,0.0002443418,0.007305884,0.8841817,0.04155255,0.0005289469,0.000462616,0.001262866,0.000502122,0.007837685,0.05578182,0.0001295622],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0008505097,0.9944786,0.002451607,0.001241961,0.0001584284,0.0001697033,0.0002571621,0.00002384873,0.0003683316],"genre_scores_gemma":[0.042722,0.944599,0.008653037,0.002033841,0.0004792181,0.0008027459,0.0005010793,0.00004345558,0.0001654968],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9612992,"threshold_uncertainty_score":0.2046719,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03279360825787257,"score_gpt":0.348339967714137,"score_spread":0.3155463594562644,"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."}}