{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001990116,0.0004237704,0.001565422,0.0002990075,0.0004038791,0.00001579081,0.0003676849,0.0002039802,0.0002470114],"category_scores_gemma":[0.0004958561,0.0004090518,0.0002463635,0.000392739,0.001188237,0.0001187734,0.0002586949,0.001030352,7.434185e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001182447,"about_ca_system_score_gemma":0.0001765948,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002233865,"about_ca_topic_score_gemma":0.000002830307,"domain_scores_codex":[0.9953369,0.001607801,0.001163633,0.0007024606,0.0004911359,0.0006980165],"domain_scores_gemma":[0.9963875,0.002062691,0.0006766558,0.0004719339,0.0001378199,0.0002633971],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002960049,0.0002091108,0.712513,0.0004439218,0.0009249517,0.0002123698,0.001841775,0.07362244,0.201355,0.00009189999,0.002707913,0.003117593],"study_design_scores_gemma":[0.005308941,0.02323372,0.5829815,0.0002722518,0.0010184,0.01415445,0.0008626189,0.3674403,0.0002328451,0.00296306,0.0006835649,0.0008483466],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9908114,0.005991501,0.001081783,0.0006123827,0.0004708881,0.0006914749,0.0001247182,0.00008709244,0.0001287394],"genre_scores_gemma":[0.9870354,0.0001148118,0.01160295,0.0002590002,0.0003293804,0.00008825975,0.0004171688,0.00006238839,0.00009065718],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2938179,"threshold_uncertainty_score":0.9998361,"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."}}