{"id":"W1979134293","doi":"10.1002/sim.3310","title":"Estimating error rates in the classification of paired organs","year":2008,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Estimator; Resampling; Statistics; Cross-validation; Word error rate; Computer science; Sample size determination; Contrast (vision); Autocorrelation; Artificial intelligence; Variance (accounting); Pattern recognition (psychology); Mathematics","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":[0.0006683433,0.000079094,0.0002942127,0.0001695612,0.00003509333,0.00000168562,0.00008194541,0.00002638579,0.00006341385],"category_scores_gemma":[0.002098657,0.00004849041,0.00001574317,0.0005277322,0.0002932479,0.00001782898,0.000006897925,0.0001930913,0.000004173576],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002902215,"about_ca_system_score_gemma":0.00005438083,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003505763,"about_ca_topic_score_gemma":0.0000528634,"domain_scores_codex":[0.9989337,0.00009221763,0.0004167414,0.0001269628,0.0003076874,0.0001226477],"domain_scores_gemma":[0.9991157,0.0004341126,0.0001186266,0.0002004535,0.0001008278,0.00003028308],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001156941,0.0003722575,0.9245659,0.0004584159,0.00005402612,0.0007042914,0.01932952,0.0002141066,0.01903023,0.00410021,0.01679671,0.0142587],"study_design_scores_gemma":[0.00143525,0.0002274622,0.8527178,0.0005502552,0.0001137729,0.00009827538,0.003936165,0.138543,0.0001894138,0.001967433,0.0001440793,0.00007714375],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.906355,0.0003743208,0.08013564,0.009707894,0.0001169974,0.0002752367,0.00001484759,0.00002061279,0.002999481],"genre_scores_gemma":[0.9626355,0.00006020802,0.0367013,0.0002668722,0.00008375329,0.00000705886,0.00007218427,0.000007920579,0.0001651669],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1383289,"threshold_uncertainty_score":0.251244,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06777729135711709,"score_gpt":0.380192824480855,"score_spread":0.3124155331237379,"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."}}