{"id":"W6976607751","doi":"10.60692/nwy6r-ht733","title":"Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation: The M&Ms Challenge","year":2021,"lang":"en","type":"article","venue":"Greater South Information System","topic":"Cardiac Imaging and Diagnostics","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Canadian VIGOUR Centre; University of Alberta","funders":"","keywords":"Generalizability theory; Deep learning; Benchmarking; Cardiac magnetic resonance; Segmentation; Scanner; Cardiac imaging; Homogeneous","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.009583108,0.002032739,0.002355203,0.001936698,0.001219844,0.002324858,0.002274627,0.005494705,0.001310052],"category_scores_gemma":[0.01424154,0.0008180478,0.001842205,0.002370662,0.001556245,0.001352166,0.002772916,0.002297628,0.001073321],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001515439,"about_ca_system_score_gemma":0.002559771,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009611794,"about_ca_topic_score_gemma":0.01571966,"domain_scores_codex":[0.9941949,0.002041309,0.0003878852,0.001907285,0.0009784163,0.0004902183],"domain_scores_gemma":[0.9889107,0.005604587,0.0007662656,0.002242007,0.001253242,0.001223188],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003430404,0.0009938958,0.05374991,0.003319412,0.001824566,0.004976782,0.001853851,0.1364917,0.02649414,0.006554815,0.2306717,0.5296389],"study_design_scores_gemma":[0.0008314512,0.001164655,0.09760746,0.0008529806,0.001073377,0.01448364,0.003453208,0.6270934,0.04783582,0.04727024,0.1577997,0.0005340948],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6396429,0.02467906,0.25646,0.03023263,0.003828098,0.0009397618,0.02607372,0.008338506,0.009805453],"genre_scores_gemma":[0.7348372,0.00318804,0.2084398,0.004668059,0.001925967,0.0003605668,0.03966361,0.001639585,0.005277163],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009611794,"threshold_uncertainty_score":0.05068088,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04030624901725267,"score_gpt":0.2527021452858726,"score_spread":0.2123958962686199,"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."}}