{"id":"W4413426693","doi":"10.1213/xaa.0000000000002040","title":"From Consensus to Standardization: Evaluating Deep Learning for Nerve Block Segmentation in Ultrasound Imaging","year":2025,"lang":"en","type":"article","venue":"A&A Practice","topic":"Anesthesia and Pain Management","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Standardization; Segmentation; Consensus conference; Ultrasound imaging; Block (permutation group theory); Ultrasound; Artificial intelligence; Computer science; Medicine; Radiology; Internal medicine","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":[0.02654983,0.002594752,0.001777946,0.00372645,0.001249114,0.002538585,0.003218119,0.00434994,0.003331687],"category_scores_gemma":[0.05892406,0.0007709684,0.001883363,0.001609624,0.002301957,0.002768933,0.00507177,0.00290839,0.001572047],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002347441,"about_ca_system_score_gemma":0.00307651,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008590763,"about_ca_topic_score_gemma":0.008299628,"domain_scores_codex":[0.9885494,0.004306386,0.001104533,0.003252054,0.00218403,0.0006035385],"domain_scores_gemma":[0.972291,0.01388226,0.002416475,0.003472458,0.00675823,0.001179488],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00573048,0.000988915,0.04255633,0.002089576,0.001676978,0.000314832,0.0008670393,0.3719367,0.006167424,0.00307129,0.01746988,0.5471305],"study_design_scores_gemma":[0.0004203891,0.00221344,0.008666112,0.0004720537,0.0003611469,0.000303577,0.0005002515,0.9652626,0.01181071,0.005835699,0.004064189,0.0000898254],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6159673,0.0135532,0.3330832,0.001955841,0.001691577,0.002428103,0.00332145,0.01276332,0.01523596],"genre_scores_gemma":[0.8806825,0.001179387,0.1063821,0.000685673,0.0001845451,0.0005400539,0.006997654,0.0007571743,0.002590893],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02654983,"threshold_uncertainty_score":0.1404106,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02249361048925378,"score_gpt":0.3785898669824092,"score_spread":0.3560962564931554,"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."}}