{"id":"W4392383104","doi":"10.32920/25336324","title":"Deep Learning and Explainable AI Methods for Surgical Skills Assessment","year":2024,"lang":"en","type":"preprint","venue":"","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University","funders":"Sunnybrook Research Institute","keywords":"Knot tying; Computer science; Tying; Artificial intelligence; Deep learning; Task (project management); Machine learning; Modal; Black box; Engineering; Medicine; Surgery; Systems engineering","routes":{"ca_aff":true,"ca_fund":true,"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":[0.001197071,0.001005926,0.0004473875,0.0009329129,0.0002253098,0.001012216,0.001181595,0.001142267,0.003573437],"category_scores_gemma":[0.004701328,0.0003331797,0.0008939629,0.0007386411,0.0005851832,0.001213145,0.0009762686,0.00213807,0.0006763618],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001273692,"about_ca_system_score_gemma":0.0007875054,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005449887,"about_ca_topic_score_gemma":0.007163538,"domain_scores_codex":[0.9994907,0.0001956047,0.00002600498,0.0001420338,0.0000967029,0.00004891984],"domain_scores_gemma":[0.9981646,0.001226798,0.0001862947,0.0001790345,0.0001815162,0.0000617986],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001516286,0.0001530572,0.002885202,0.0002810686,0.0001858921,0.0001003532,0.0001487942,0.5241964,0.004502654,0.02200139,0.008494517,0.436899],"study_design_scores_gemma":[0.0000063327,0.00002062247,0.0005691692,0.00001911983,0.000009243763,0.00001247619,0.000009865596,0.9788997,0.0008524033,0.01835783,0.001235961,0.000007254482],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02426014,0.002207614,0.9666981,0.001312979,0.000143131,0.00007553707,0.0008768376,0.001858995,0.002566684],"genre_scores_gemma":[0.7432564,0.001682304,0.243521,0.0004293062,0.0002378046,0.0002603437,0.002422537,0.0002480767,0.007942162],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005449887,"threshold_uncertainty_score":0.01195437,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03815116142575749,"score_gpt":0.4673780003506187,"score_spread":0.4292268389248612,"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."}}