{"id":"W4410622442","doi":"10.1016/j.jvs.2025.03.132","title":"Evaluating VASC.AI: A Vascular Surgery-Specific AI Model Excelling in Certification Exam-Level Questions","year":2025,"lang":"en","type":"article","venue":"Journal of Vascular Surgery","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Medicine; Certification; Medical physics; Management","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":[0.008258169,0.0006353426,0.0006148841,0.001860647,0.000790488,0.003148063,0.001178656,0.001070564,0.009002412],"category_scores_gemma":[0.03562484,0.0002179877,0.001411479,0.000925904,0.0005343696,0.002434469,0.002203834,0.001632018,0.002521169],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002517077,"about_ca_system_score_gemma":0.004360229,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006630539,"about_ca_topic_score_gemma":0.01686778,"domain_scores_codex":[0.9947942,0.001919209,0.0005593185,0.0005479479,0.001446156,0.0007331804],"domain_scores_gemma":[0.9724633,0.01018538,0.005271462,0.001412893,0.006032072,0.004634898],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00097508,0.003474935,0.8404114,0.0002050234,0.000214521,0.00008211271,0.0007102147,0.002570946,0.0007935253,0.003210083,0.01919585,0.1281563],"study_design_scores_gemma":[0.0004736979,0.003164314,0.9098116,0.0003604302,0.0005536849,0.000503845,0.002353265,0.03947484,0.004326319,0.006456044,0.0323789,0.0001431572],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9110983,0.000410028,0.01115596,0.004121547,0.0003371657,0.002170958,0.004371192,0.001012091,0.06532268],"genre_scores_gemma":[0.9660532,0.0002101437,0.02243237,0.0009151951,0.0001047482,0.001284648,0.003380381,0.0001028019,0.005516515],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009002412,"threshold_uncertainty_score":0.04367387,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1138216107084663,"score_gpt":0.3515203703478591,"score_spread":0.2376987596393927,"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."}}