{"id":"W4415052657","doi":"10.2196/71994","title":"Evaluation of Machine Learning Model Performance in Diabetic Foot Ulcer: Retrospective Cohort Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Diabetic Foot Ulcer Assessment and Management","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Haaglanden Medisch Centrum","keywords":"Brier score; Retrospective cohort study; Calibration; Support vector machine; Feature (linguistics); Clinical trial; Receiver operating characteristic; Relevance (law)","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.01674996,0.0005326005,0.0008308249,0.001367651,0.0003850625,0.001160492,0.0008168891,0.000822102,0.00158283],"category_scores_gemma":[0.02644842,0.0002678465,0.001862003,0.0009087212,0.0003320064,0.0006872431,0.0006784125,0.001091349,0.0003894706],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004358192,"about_ca_system_score_gemma":0.000508977,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002526026,"about_ca_topic_score_gemma":0.001458086,"domain_scores_codex":[0.9965156,0.001482173,0.0004468577,0.000836436,0.0005298314,0.0001890361],"domain_scores_gemma":[0.9789586,0.01147964,0.002854341,0.00364934,0.002486912,0.0005711793],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001247078,0.0001586329,0.9894518,0.00003696753,0.0008910128,0.0001013244,0.0000852471,0.001779398,0.0001497926,0.0000678233,0.0004533698,0.005577422],"study_design_scores_gemma":[0.0001787635,0.002730612,0.9333978,0.0000932531,0.001733119,0.001228987,0.0005708919,0.05692027,0.0009467326,0.000414582,0.001717702,0.00006725011],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9963855,0.0004228903,0.001927187,0.00005023187,0.00002244557,0.00003250731,0.0009690486,0.00002293429,0.0001672543],"genre_scores_gemma":[0.9974425,0.000120515,0.0006482244,0.00001896111,0.00001609572,0.00003262977,0.001598694,0.00001074905,0.000111609],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01674996,"threshold_uncertainty_score":0.08858335,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01823773697579406,"score_gpt":0.3344674947777611,"score_spread":0.316229757801967,"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."}}