{"id":"W3108591672","doi":"10.1002/mp.14676","title":"Toward data‐efficient learning: A benchmark for COVID‐19 CT lung and infection segmentation","year":2020,"lang":"en","type":"article","venue":"Medical Physics","topic":"COVID-19 diagnosis using AI","field":"Medicine","cited_by":251,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"National Natural Science Foundation of China","keywords":"Benchmark (surveying); Segmentation; Medical imaging; Image segmentation; Deep learning; Computed tomography; Pattern recognition (psychology)","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":[0.005856964,0.003503528,0.001270211,0.002858442,0.001176997,0.002815521,0.003963072,0.003849298,0.002578902],"category_scores_gemma":[0.01646421,0.0007904412,0.001575537,0.002227638,0.001597749,0.002292105,0.002918639,0.002430763,0.002035328],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002953506,"about_ca_system_score_gemma":0.002838758,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02045947,"about_ca_topic_score_gemma":0.02157206,"domain_scores_codex":[0.9962388,0.0009652585,0.0003817072,0.001109898,0.0009161633,0.000388169],"domain_scores_gemma":[0.9948034,0.002034008,0.0004433087,0.0009339675,0.00143245,0.000352919],"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":[0.001719559,0.00128071,0.01893253,0.002390171,0.0008338661,0.0009867596,0.0003374132,0.5530443,0.01111346,0.004883755,0.09417595,0.3103016],"study_design_scores_gemma":[0.0001936683,0.0005295565,0.00464339,0.0002553362,0.0000990577,0.0005913186,0.0002092964,0.9510076,0.02256317,0.007272054,0.01256251,0.00007299503],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5178487,0.020044,0.3306588,0.006837989,0.002499323,0.002247501,0.04193642,0.05390986,0.0240175],"genre_scores_gemma":[0.5815969,0.002825677,0.309722,0.001873833,0.0003125106,0.0009709757,0.09480206,0.002436574,0.005459616],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02045947,"threshold_uncertainty_score":0.04068077,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08268855983109663,"score_gpt":0.3785273690815479,"score_spread":0.2958388092504513,"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."}}