{"id":"W1993947467","doi":"10.1016/s1076-6332(03)00671-8","title":"Statistical validation of image segmentation quality based on a spatial overlap index1","year":2004,"lang":"en","type":"article","venue":"Academic Radiology","topic":"Medical Image Segmentation Techniques","field":"Computer Science","cited_by":1857,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto","funders":"National Center for Research Resources; U.S. National Library of Medicine; National Cancer Institute; National Institute on Aging; Agency for Healthcare Research and Quality","keywords":"Reproducibility; Segmentation; Nuclear medicine; Artificial intelligence; Medicine; Pattern recognition (psychology); Computer science; Image quality; Metric (unit); Magnetic resonance imaging; Radiology; Mathematics; Statistics; Image (mathematics)","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":[],"consensus_categories":[],"category_scores_codex":[0.01806311,0.0008461572,0.001117804,0.005617236,0.0007524987,0.001765019,0.001150828,0.001915653,0.001176202],"category_scores_gemma":[0.04891331,0.0004575863,0.001146558,0.00255894,0.002006807,0.001233372,0.001203945,0.0007418826,0.0005730251],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007841255,"about_ca_system_score_gemma":0.0009362088,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002320225,"about_ca_topic_score_gemma":0.002287172,"domain_scores_codex":[0.9908131,0.00319036,0.001031922,0.001538287,0.003087645,0.0003387058],"domain_scores_gemma":[0.881269,0.07949261,0.01001838,0.009581042,0.01844204,0.001196975],"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":[0.01204357,0.0008161356,0.5804407,0.0008818599,0.003003138,0.0003566467,0.0009111574,0.07391173,0.1384893,0.001770646,0.001677143,0.185698],"study_design_scores_gemma":[0.0002288665,0.002936658,0.4214549,0.00007694531,0.001097816,0.001562504,0.0003100157,0.461032,0.1086076,0.001517382,0.0009668727,0.0002084646],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8506214,0.0008717004,0.1447248,0.00009252181,0.00005085747,0.0001862105,0.001136111,0.001324345,0.0009919824],"genre_scores_gemma":[0.9623998,0.0001477742,0.03442411,0.00003767697,0.00004132977,0.00009795141,0.002362302,0.0002023566,0.0002867118],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01806311,"threshold_uncertainty_score":0.09552801,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02476122096801544,"score_gpt":0.3556104642776692,"score_spread":0.3308492433096538,"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."}}