{"id":"W4360981867","doi":"10.1109/isbi53787.2023.10230482","title":"Reproducibility of Tumor Segmentation Outcomes with a Deep Learning Model","year":2023,"lang":"en","type":"article","venue":"","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"LabEx PRIMES; Agence Nationale de la Recherche","keywords":"Reproducibility; Segmentation; Pipeline (software); Computer science; Benchmark (surveying); Stability (learning theory); Dice; Deep learning; Artificial intelligence; Hausdorff distance; Range (aeronautics); Image segmentation; Pattern recognition (psychology); Machine learning; Statistics; Mathematics; Cartography; Engineering; Geography","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009092501,0.0009056174,0.000791533,0.001094343,0.0005674217,0.002252727,0.001234987,0.001551542,0.001047952],"category_scores_gemma":[0.04169729,0.0005383326,0.0008712768,0.0006985941,0.001650062,0.00129647,0.001739972,0.001573252,0.0004258757],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001563743,"about_ca_system_score_gemma":0.001375973,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003957381,"about_ca_topic_score_gemma":0.003194086,"domain_scores_codex":[0.9956786,0.00115484,0.0004630862,0.001116427,0.001344994,0.0002420164],"domain_scores_gemma":[0.9791507,0.01157414,0.002047401,0.004162933,0.002615792,0.0004490492],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.003539026,0.0002394176,0.03877243,0.0006379323,0.0006544212,0.0005446054,0.0006716855,0.7962089,0.07123423,0.003171691,0.003392063,0.0809337],"study_design_scores_gemma":[0.00007406881,0.000556607,0.01489381,0.00007872037,0.0001361625,0.0002990065,0.0001691076,0.8841002,0.09136219,0.006367913,0.001844658,0.0001175214],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.757171,0.001476785,0.2319005,0.001041395,0.0003668235,0.000169218,0.001109364,0.004257719,0.002507209],"genre_scores_gemma":[0.9657671,0.0001579894,0.03158325,0.0001623512,0.00002538116,0.00006351776,0.001052968,0.000482276,0.000705204],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9909075,"threshold_uncertainty_score":0.04808629,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0199353249628272,"score_gpt":0.3198180548936521,"score_spread":0.2998827299308249,"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."}}