{"id":"W2921108820","doi":"10.1117/12.2512819","title":"Reliability of computer-aided diagnosis tools with multi-center MR datasets: impact of training protocol","year":2019,"lang":"en","type":"article","venue":"Medical Imaging 2019: Computer-Aided Diagnosis","topic":"AI in cancer detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Foothills Medical Centre; Ontario Brain Institute; Calgary Laboratory Services; University of Calgary","funders":"","keywords":"Computer science; Protocol (science); Cross-validation; Artificial intelligence; Classifier (UML); Data set; CAD; Computer-aided diagnosis; Reliability (semiconductor); Data mining; Data collection; Set (abstract data type); Pattern recognition (psychology); Machine learning; Medicine; Statistics; Mathematics; Pathology","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":[0.07626033,0.001218496,0.0009842344,0.0009848544,0.0007323728,0.00161279,0.001869224,0.001894557,0.001049782],"category_scores_gemma":[0.1972952,0.0007680096,0.001090612,0.001168725,0.002409482,0.001461022,0.002049983,0.001211372,0.0008048053],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009352752,"about_ca_system_score_gemma":0.001163514,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002958578,"about_ca_topic_score_gemma":0.002299336,"domain_scores_codex":[0.9380031,0.03834768,0.006684141,0.01033502,0.005844273,0.0007856653],"domain_scores_gemma":[0.7251103,0.1975679,0.01400427,0.03696397,0.02532212,0.001031458],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.05245955,0.004369586,0.5038283,0.002174421,0.005766558,0.0006820921,0.004802055,0.1137549,0.05643379,0.001165922,0.006606608,0.2479563],"study_design_scores_gemma":[0.001757894,0.01936116,0.6768566,0.0005971571,0.002882081,0.001761425,0.001495088,0.1949067,0.0890304,0.002629625,0.008155608,0.0005663036],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9446042,0.001030505,0.04897362,0.0002452437,0.0002226319,0.001271114,0.001209774,0.0008197768,0.00162321],"genre_scores_gemma":[0.9803048,0.0001180891,0.01545649,0.0001567796,0.00003419862,0.000947241,0.002304555,0.000214081,0.0004637612],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07626033,"threshold_uncertainty_score":0.403308,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0261076682686942,"score_gpt":0.3195067222023485,"score_spread":0.2933990539336543,"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."}}