{"id":"W4385190584","doi":"10.1016/j.neuroimage.2023.120289","title":"ROOD-MRI: Benchmarking the robustness of deep learning segmentation models to out-of-distribution and corrupted data in MRI","year":2023,"lang":"en","type":"article","venue":"NeuroImage","topic":"Advanced Neural Network Applications","field":"Computer Science","cited_by":32,"is_retracted":false,"has_abstract":true,"ca_institutions":"Baycrest Hospital; Heart and Stroke Foundation; University of Toronto; Western University; Sunnybrook Health Science Centre","funders":"Natural Sciences and Engineering Research Council of Canada; Government of Ontario; Canada Research Chairs; Compute Canada; Temerty Family Foundation; Ontario Brain Institute; University of Ottawa","keywords":"Benchmarking; Robustness (evolution); Segmentation; Computer science; Artificial intelligence; Benchmark (surveying); Deep learning; Pattern recognition (psychology); Machine learning; Data mining; Cartography","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":[0.006589129,0.003235955,0.000936459,0.001926447,0.0006626488,0.001689869,0.003432469,0.002432959,0.001576929],"category_scores_gemma":[0.01654244,0.0007751794,0.001543624,0.00140142,0.001737526,0.001813998,0.002746388,0.002605245,0.001091416],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00154839,"about_ca_system_score_gemma":0.001906097,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01158796,"about_ca_topic_score_gemma":0.01621116,"domain_scores_codex":[0.996946,0.000964138,0.0003004486,0.000850482,0.0007042381,0.0002345918],"domain_scores_gemma":[0.9951018,0.002077495,0.0004902838,0.001131277,0.0008981043,0.000300996],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001582288,0.0007829706,0.01663481,0.001630563,0.001461077,0.0004800706,0.0002636614,0.7645551,0.02055169,0.004195505,0.03204885,0.1558134],"study_design_scores_gemma":[0.000178949,0.0006411398,0.005468519,0.000147508,0.0001198807,0.0002952774,0.0001071108,0.9493837,0.03237588,0.005345189,0.005852311,0.00008461921],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6249418,0.008109783,0.3003478,0.002480459,0.001698454,0.0009011098,0.01557114,0.03612269,0.00982682],"genre_scores_gemma":[0.7234749,0.001760374,0.2238338,0.001486509,0.0001987315,0.0006412193,0.04170617,0.003060968,0.003837221],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01158796,"threshold_uncertainty_score":0.03484708,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06557973590540603,"score_gpt":0.3127212765602763,"score_spread":0.2471415406548702,"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."}}