{"id":"W3211220493","doi":"10.1093/bjs/znab361.022","title":"SP2.1.1Continuous Monitoring and Assessment of Surgical Technical Skills Using Deep Learning","year":2021,"lang":"en","type":"article","venue":"British journal of surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Medicine; Virtual reality; Forceps; Medical physics; Artificial intelligence; Simulation; Surgery; Computer science","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":[0.001023912,0.00008729653,0.0006176902,0.000117509,0.00007153034,0.00004868971,0.00002843799,0.00009624348,0.0001699188],"category_scores_gemma":[0.0006237112,0.00009484825,0.0002539815,0.0002283445,0.00006654585,0.0001208504,0.0000378732,0.0005610334,2.764785e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004714956,"about_ca_system_score_gemma":0.0001818848,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001034675,"about_ca_topic_score_gemma":7.313391e-7,"domain_scores_codex":[0.9983898,0.0001323938,0.0007276595,0.0001303479,0.0004291718,0.0001906642],"domain_scores_gemma":[0.9980397,0.0009966858,0.0003218405,0.00005857543,0.000380657,0.0002025766],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004582184,0.0002237553,0.7028256,0.00005679297,0.00008879851,0.01558064,0.00002428893,0.0002537442,0.002105191,0.00001280849,0.000008652907,0.2787739],"study_design_scores_gemma":[0.002408935,0.00006168053,0.9357041,0.002634807,0.0001309722,0.05280827,0.0003210912,0.0006481085,0.001135799,0.00002597872,0.003973226,0.0001470696],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9938389,0.004295354,0.0003003961,0.00006969539,0.0002589605,0.0000326679,7.287027e-7,0.00001378061,0.001189515],"genre_scores_gemma":[0.9928533,0.002402732,0.004305498,0.00001458911,0.0003552733,3.499206e-7,0.000002183522,0.00001634715,0.00004976229],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2786268,"threshold_uncertainty_score":0.3867799,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04037546870545888,"score_gpt":0.3392891681418409,"score_spread":0.298913699436382,"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."}}