{"id":"W4387978272","doi":"10.1016/j.media.2023.103015","title":"Do we really need dice? The hidden region-size biases of segmentation losses","year":2023,"lang":"en","type":"article","venue":"Medical Image Analysis","topic":"Advanced Neural Network Applications","field":"Computer Science","cited_by":45,"is_retracted":false,"has_abstract":false,"ca_institutions":"Centre Hospitalier de l’Université de Montréal; École de Technologie Supérieure","funders":"H2020 Marie Skłodowska-Curie Actions; Fonds de recherche du Québec – Nature et technologies; HORIZON EUROPE Marie Sklodowska-Curie Actions; Natural Sciences and Engineering Research Council of Canada","keywords":"Dice; Ground truth; Segmentation; Computer science; Artificial intelligence; Perspective (graphical); Matching (statistics); Upper and lower bounds; Pattern recognition (psychology); Mathematics; Statistics","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":[0.007381503,0.0007487925,0.001165991,0.0006066736,0.0004963845,0.002141116,0.001564766,0.002758909,0.004481582],"category_scores_gemma":[0.05124316,0.0007228667,0.0004838147,0.0005206359,0.00260403,0.01056027,0.001713849,0.003365035,0.0009088208],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001408211,"about_ca_system_score_gemma":0.0007315107,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002067599,"about_ca_topic_score_gemma":0.00234376,"domain_scores_codex":[0.9988607,0.0003793328,0.00005555468,0.000273026,0.000313416,0.0001178704],"domain_scores_gemma":[0.9869655,0.008617098,0.0007921619,0.001844373,0.001448802,0.0003320399],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0008292783,0.0001676321,0.01096189,0.0006016698,0.0002835641,0.0003319455,0.0005384925,0.1521687,0.02217549,0.3156475,0.02555146,0.4707423],"study_design_scores_gemma":[0.00007891837,0.000131164,0.006534777,0.0001708326,0.00006765222,0.0006154568,0.0001028554,0.5580589,0.01583408,0.4118278,0.006486932,0.00009063061],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1480573,0.005939152,0.8046148,0.02827825,0.0008276457,0.00008325154,0.0004155905,0.0008763261,0.0109076],"genre_scores_gemma":[0.8534451,0.002944964,0.1274174,0.003763761,0.001011952,0.00008862311,0.0003732883,0.0007482515,0.01020663],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007381503,"threshold_uncertainty_score":0.03903759,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03027854292138389,"score_gpt":0.3202295275761869,"score_spread":0.289950984654803,"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."}}