{"id":"W134925056","doi":"","title":"A web-based test of residents' skills in diagnostic radiology.","year":2006,"lang":"en","type":"article","venue":"PubMed","topic":"Radiology practices and education","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Medicine; Medical diagnosis; Test (biology); Medical physics; Radiology; Reliability (semiconductor); Diagnostic accuracy; Web application; World Wide Web; 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":[],"consensus_categories":[],"category_scores_codex":[0.0004246819,0.00006699052,0.0001971561,0.0001520726,0.00001444306,0.000003461567,0.00005234097,0.00008705398,0.00003119119],"category_scores_gemma":[0.005603407,0.00005874563,0.00003523879,0.0001847175,0.00007618107,0.00005105717,0.000006358473,0.0001152313,0.00001180446],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000646128,"about_ca_system_score_gemma":0.00008267524,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000537357,"about_ca_topic_score_gemma":0.000098745,"domain_scores_codex":[0.9992229,0.00005808187,0.000269909,0.0001452777,0.00007651942,0.0002272795],"domain_scores_gemma":[0.9974375,0.002177275,0.0001221847,0.0001781356,0.00003365905,0.00005126338],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00004140293,0.0004653235,0.9874071,0.0000285446,0.000007065674,0.00002484134,0.00001885478,0.00003835449,0.0002021526,0.00003319331,0.006691162,0.005041936],"study_design_scores_gemma":[0.001535111,0.00004887839,0.995873,0.00002029347,0.0000429401,0.00003955936,0.00001428352,0.0000469595,0.00055014,0.0001432581,0.001635297,0.00005023336],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9916657,0.0005440988,0.000007046932,0.003489909,0.000196166,0.0005125589,0.000003394066,0.00001922196,0.003561905],"genre_scores_gemma":[0.9979253,0.00006234612,0.000204197,0.0003379676,0.0001790268,0.0004961452,0.00002760304,0.000008304154,0.0007590856],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008465878,"threshold_uncertainty_score":0.6708206,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01112869925680041,"score_gpt":0.2551901714978926,"score_spread":0.2440614722410922,"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."}}