{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001089067,0.00007479128,0.0002253082,0.00008903203,0.00004225147,0.000006196001,0.00004019096,0.0000164732,0.00005568775],"category_scores_gemma":[0.001209219,0.00004855634,0.00004681509,0.0002430887,0.00006052411,0.00004387845,0.00002809138,0.0002063363,0.00001448871],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002162349,"about_ca_system_score_gemma":0.00004107479,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007122935,"about_ca_topic_score_gemma":0.000004106063,"domain_scores_codex":[0.9989624,0.00002963078,0.0001997587,0.0004168922,0.0002503569,0.0001409855],"domain_scores_gemma":[0.9992918,0.00006762692,0.00007066099,0.000437182,0.00006206449,0.00007064641],"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":[0.00009396332,0.00007272876,0.9204172,0.0001566949,0.00006725436,0.00002870514,0.000963892,0.04238548,0.01194567,0.0002464773,0.0002443675,0.02337757],"study_design_scores_gemma":[0.0009940179,0.0001460728,0.1743038,0.00004103013,0.00005403866,0.00001955168,0.0005641334,0.821211,0.00231901,0.0002165397,0.00005926742,0.00007155623],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9703681,0.00001269264,0.02395503,0.002355336,0.00002771602,0.000178718,3.031516e-7,0.0001885357,0.002913584],"genre_scores_gemma":[0.9725935,0.000008355732,0.0225412,0.0003802484,0.00001982834,0.000009688586,0.00001735064,0.00001618477,0.00441361],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7788255,"threshold_uncertainty_score":0.198007,"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."}}