{"id":"W3110759004","doi":"10.1002/alz.041150","title":"Reliability assessment of tissue classification algorithms for multi‐center and multi‐scanner data","year":2020,"lang":"en","type":"article","venue":"Alzheimer s & Dementia","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval","funders":"","keywords":"Kappa; Segmentation; Artificial intelligence; Scanner; Pattern recognition (psychology); Dice; Computer science; Nuclear medicine; Mathematics; Cartography; Algorithm; Medicine; Statistics; Geography; Geometry","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":[0.07422674,0.001425505,0.001467137,0.003535522,0.001918286,0.002966193,0.002663461,0.001765791,0.002318653],"category_scores_gemma":[0.1886759,0.001014481,0.002262663,0.002664685,0.003067353,0.002039673,0.003958585,0.001761552,0.00149964],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009816188,"about_ca_system_score_gemma":0.001221671,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002339038,"about_ca_topic_score_gemma":0.003125426,"domain_scores_codex":[0.9504036,0.02108905,0.005476498,0.0137238,0.008374772,0.0009322275],"domain_scores_gemma":[0.7340831,0.1658472,0.02167198,0.04482722,0.03194103,0.001629378],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007848407,0.0005495721,0.5236325,0.002737505,0.01234961,0.0004874793,0.004310891,0.06933256,0.04238175,0.004857346,0.01252054,0.3189918],"study_design_scores_gemma":[0.0005447243,0.002260104,0.5013742,0.0007178802,0.003547779,0.002841495,0.001288354,0.3920436,0.05260415,0.02457969,0.01763912,0.0005588867],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5747018,0.003369198,0.4103335,0.0006066023,0.000475405,0.0008437973,0.00246498,0.003939172,0.003265499],"genre_scores_gemma":[0.8790394,0.000254987,0.1141217,0.0002228563,0.0001362848,0.0007893153,0.003128338,0.001741303,0.0005658116],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07422674,"threshold_uncertainty_score":0.3925532,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1239308809581636,"score_gpt":0.408638995087604,"score_spread":0.2847081141294404,"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."}}