{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001278681,0.00008545345,0.0001385006,0.0001035498,0.0001113812,0.00005457427,0.00003667212,0.00002641061,0.00003624052],"category_scores_gemma":[0.004900487,0.00008999572,0.00002930632,0.0002620691,0.00001233821,0.0001052734,0.00001427371,0.000121055,0.00001414667],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001351146,"about_ca_system_score_gemma":0.00007434208,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002249028,"about_ca_topic_score_gemma":0.00002066685,"domain_scores_codex":[0.9989602,0.0002383447,0.0002325793,0.0002226649,0.0002012724,0.0001449436],"domain_scores_gemma":[0.9966054,0.002927665,0.0001081499,0.000126058,0.000195165,0.00003754586],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.006607158,0.0007480873,0.4366812,0.0008553299,0.001058571,0.001495188,0.03070199,0.1412599,0.04418756,0.003984873,0.02836333,0.3040568],"study_design_scores_gemma":[0.0138644,0.0009397723,0.0764611,0.001332694,0.002824238,0.0003833743,0.1184337,0.1269535,0.005889299,0.001955048,0.6501675,0.0007954058],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6119183,0.0008653603,0.3180265,0.0530506,0.0002355295,0.002828141,0.000002423372,0.0001348774,0.01293835],"genre_scores_gemma":[0.9204036,0.00001843381,0.06990272,0.008116403,0.0001220997,0.0000917292,0.00006615713,0.00001602222,0.001262857],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6218042,"threshold_uncertainty_score":0.5866694,"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."}}