{"id":"W3004310692","doi":"10.1101/2020.01.28.922971","title":"Reliability Assessment of Tissue Classification Algorithms for Multi-Center and Multi-Scanner Data","year":2020,"lang":"en","type":"preprint","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Medical Image Segmentation Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"Consortium canadien en neurodégénérescence associée au vieillissement","keywords":"Scanner; Segmentation; Robustness (evolution); Artificial intelligence; Sample size determination; Computer science; Pattern recognition (psychology); Brain size; Statistics; Mathematics; Nuclear medicine; Medicine; Magnetic resonance imaging; Biology; Radiology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.08322822,0.001382579,0.001473412,0.00449623,0.001646858,0.00300188,0.0026228,0.001624829,0.001730572],"category_scores_gemma":[0.1819398,0.0008889795,0.002686228,0.003159436,0.002330079,0.001839488,0.003304919,0.001621948,0.001147861],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001116544,"about_ca_system_score_gemma":0.001350371,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003540065,"about_ca_topic_score_gemma":0.003971183,"domain_scores_codex":[0.9488086,0.02546094,0.005591439,0.01213764,0.007081945,0.0009194482],"domain_scores_gemma":[0.7039417,0.1918935,0.02516919,0.04440682,0.03293188,0.001656911],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005965652,0.0004602442,0.6031871,0.00183514,0.01282074,0.0004440796,0.003058659,0.08368721,0.02217028,0.002546066,0.009259582,0.2545653],"study_design_scores_gemma":[0.0004475568,0.001485382,0.4341431,0.0005155495,0.00274179,0.001990748,0.001136121,0.5021066,0.03107918,0.01385093,0.01006458,0.0004384384],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.6297755,0.00264423,0.3567029,0.0006080816,0.0004457885,0.0008099052,0.002779276,0.004019124,0.002215314],"genre_scores_gemma":[0.8777324,0.00021201,0.1165989,0.0001843955,0.0001159972,0.0005187498,0.003219036,0.00101232,0.000406088],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.08322822,"threshold_uncertainty_score":0.4401581,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1129960362770045,"score_gpt":0.3626591045380474,"score_spread":0.249663068261043,"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."}}