{"id":"W2341250765","doi":"10.36834/cmej.36654","title":"Interpretation difficulty of normal versus abnormal radiographs using a pediatric example","year":2016,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Radiology practices and education","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"ca_institutions":"Hospital for Sick Children","funders":"University of Toronto","keywords":"Interpretation (philosophy); Radiography; Computer science; Natural language processing; Artificial intelligence; Medicine; Pattern recognition (psychology); Radiology; Programming language","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0006028734,0.0001028648,0.0001859814,0.0005737803,0.0001382207,0.00001855583,0.000129289,0.0001772721,0.008172091],"category_scores_gemma":[0.005926888,0.00007271631,0.0001012192,0.0003305645,0.0001160216,0.0003121736,0.000006623835,0.0003215432,0.00002034178],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003751206,"about_ca_system_score_gemma":0.02533135,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005587428,"about_ca_topic_score_gemma":0.001943649,"domain_scores_codex":[0.998647,0.0001040351,0.0004217821,0.0001470199,0.0003773762,0.0003027251],"domain_scores_gemma":[0.9955528,0.0001648381,0.0002634794,0.0001534568,0.0003612397,0.003504227],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00149746,0.0008246591,0.3828077,0.0001680496,0.0003940317,0.0001011391,0.005722868,0.00001801177,0.001014043,0.0009227501,0.1208919,0.4856374],"study_design_scores_gemma":[0.01240303,0.001806041,0.8098581,0.0009792758,0.001493093,0.01513789,0.006660128,0.002247137,0.0002515172,0.0003002094,0.1480126,0.0008509973],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.984106,0.0005243194,0.0005006907,0.009994292,0.003745323,0.0001113721,0.000004189128,0.000008065311,0.001005732],"genre_scores_gemma":[0.9948878,0.0004198002,0.001068098,0.001521064,0.001908584,0.000005088788,0.00001253495,0.00001273163,0.0001642646],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4847864,"threshold_uncertainty_score":0.9927346,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02515649206023056,"score_gpt":0.3164186960043612,"score_spread":0.2912622039441306,"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."}}