{"id":"W4402601437","doi":"10.1148/ryai.230550","title":"Assessing the Performance of Models from the 2022 RSNA Cervical Spine Fracture Detection Competition at a Level I Trauma Center","year":2024,"lang":"en","type":"article","venue":"Radiology Artificial Intelligence","topic":"Medical Imaging and Analysis","field":"Engineering","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"St. Michael's Hospital Foundation","keywords":"Cervical spine; Center (category theory); Trauma center; Medicine; Competition (biology); Internal medicine; Surgery; Retrospective cohort study","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":[0.009845598,0.004354552,0.001779771,0.001814619,0.001293429,0.002188518,0.002407242,0.002851257,0.00219952],"category_scores_gemma":[0.01386493,0.0006461741,0.002933232,0.0008293521,0.0006504584,0.00124786,0.001726603,0.003271908,0.001852689],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003149553,"about_ca_system_score_gemma":0.003709342,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.06234243,"about_ca_topic_score_gemma":0.06388316,"domain_scores_codex":[0.9968681,0.001159872,0.0002123052,0.0008157627,0.0005590661,0.0003847887],"domain_scores_gemma":[0.9929345,0.003338046,0.000294114,0.0006264034,0.002094615,0.0007121874],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.003287358,0.003017656,0.09947626,0.0005603983,0.003065683,0.0004009837,0.0002527712,0.6576335,0.004700696,0.001101579,0.05337771,0.1731253],"study_design_scores_gemma":[0.0002235129,0.001559007,0.01750866,0.00006965282,0.0003016396,0.0001673803,0.0001820906,0.9707992,0.004448851,0.001168792,0.003478931,0.0000922383],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9426369,0.003802848,0.02547152,0.00373989,0.001670379,0.0006814967,0.009095905,0.003350798,0.00955022],"genre_scores_gemma":[0.9505214,0.0004792618,0.01764186,0.0008578927,0.000235423,0.0002514398,0.02326198,0.0002958681,0.006454889],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.06234243,"threshold_uncertainty_score":0.1239591,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05115680240699588,"score_gpt":0.2940749643982224,"score_spread":0.2429181619912265,"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."}}