{"id":"W4210918265","doi":"10.1111/vru.13069","title":"Comparison of error rates between four pretrained DenseNet convolutional neural network models and 13 board‐certified veterinary radiologists when evaluating 15 labels of canine thoracic radiographs","year":2022,"lang":"en","type":"article","venue":"Veterinary Radiology & Ultrasound","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"Rogue Research (Canada); Canadian Veterinary Medical Association; Montreal Police Service","funders":"","keywords":"Medicine; Radiography; Gold standard (test); Convolutional neural network; Institutional review board; Radiology; Veterinary medicine; Artificial intelligence; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01380913,0.0007840029,0.0005990903,0.00141757,0.0003100575,0.001061641,0.0008041924,0.001117721,0.0004294139],"category_scores_gemma":[0.04012795,0.0004274584,0.0007605612,0.0005868381,0.0009159179,0.0009877891,0.001049963,0.0005093784,0.0002536504],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001028826,"about_ca_system_score_gemma":0.0005212702,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004086003,"about_ca_topic_score_gemma":0.008489204,"domain_scores_codex":[0.9940309,0.001960755,0.0009821301,0.00135982,0.001240524,0.0004258502],"domain_scores_gemma":[0.9693639,0.01773468,0.003985788,0.002581111,0.005737925,0.0005965898],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002986304,0.0001694922,0.8950014,0.0001906603,0.0007110061,0.0004667242,0.0006607149,0.01636688,0.007750094,0.0002679393,0.0009332937,0.07449546],"study_design_scores_gemma":[0.0001566892,0.002622911,0.7792504,0.0002345103,0.0008440724,0.002317308,0.001322018,0.1754743,0.03401661,0.00091945,0.002695401,0.0001464185],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9944229,0.0003254261,0.004289072,0.00006876147,0.00003515722,0.00004264455,0.0001595874,0.00005929537,0.0005972371],"genre_scores_gemma":[0.9951422,0.00009287888,0.003912701,0.00004746124,0.00001035955,0.00002338709,0.0004614256,0.00002165999,0.0002879434],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01380913,"threshold_uncertainty_score":0.07303053,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1545214343275972,"score_gpt":0.4121974365414585,"score_spread":0.2576760022138613,"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."}}