{"id":"W4391756091","doi":"10.1038/s41592-023-02151-z","title":"Metrics reloaded: recommendations for image analysis validation","year":2024,"lang":"en","type":"review","venue":"Nature Methods","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":398,"is_retracted":false,"has_abstract":false,"ca_institutions":"Sunnybrook Health Science Centre; Vector Institute; Princess Margaret Cancer Centre; Mila - Quebec Artificial Intelligence Institute; University of Toronto; University Health Network; McGill University","funders":"National Institute of General Medical Sciences; National Cancer Institute; Medical Research Council; National Institutes of Health; Masarykova Univerzita; Novo Nordisk Fonden; Universidad de Buenos Aires; University of Pennsylvania; Novo Nordisk; KU Leuven; University of Oxford; Cancer Research UK; University College London; Universitetet i Tromsø; King's College London; Oulun Yliopisto; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Institute of Neurological Disorders and Stroke; Canadian Institute for Advanced Research; Royal Academy of Engineering; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; NIH Clinical Center; Agence Nationale de la Recherche; Innosuisse - Schweizerische Agentur für Innovationsförderung; Alzheimer's Society; School of Medicine, Indiana University; European Commission; Silicon Valley Community Foundation; Broad Institute; University of Bern; Sydäntutkimussäätiö; Deutsches Krebsforschungszentrum; Natural Sciences and Engineering Research Council of Canada; Consejo Nacional de Investigaciones Científicas y Técnicas; Foundation for Cardiovascular Research; McGill University; National Science Foundation","keywords":"Computer science; Process (computing); Metric (unit); Representation (politics); Domain (mathematical analysis); Fingerprint (computing); Machine learning; Data mining; Object (grammar); Segmentation; Artificial intelligence; Convergence (economics); Selection (genetic algorithm); Image (mathematics); Algorithm; Mathematics","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":"codex-gemma-dda1882f352a","candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.003228605,0.0003268234,0.001610859,0.002225,0.0001308216,0.00008811679,0.0001592473,0.001421106,0.0002059183],"category_scores_gemma":[0.004148495,0.0002452066,0.00140877,0.006112339,0.00002964506,0.00008140023,0.00003516838,0.001770726,0.000101743],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004502676,"about_ca_system_score_gemma":0.0006625478,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004997125,"about_ca_topic_score_gemma":0.000005733987,"domain_scores_codex":[0.9973186,0.0005579079,0.001021872,0.0006067401,0.0002231105,0.0002717387],"domain_scores_gemma":[0.9958165,0.002520919,0.0003619442,0.0005870214,0.0005606253,0.0001529679],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000004374453,0.00003639204,0.000001139958,0.01128316,0.001015466,0.000001524305,0.00009296597,8.896317e-8,8.371521e-7,0.0004254731,0.03234045,0.9547981],"study_design_scores_gemma":[0.000009990201,0.00005546168,4.029921e-7,0.002057881,0.03624847,0.00001492044,0.00006287453,0.0001656839,0.0001570648,0.001778744,0.9592569,0.0001916339],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[1.574076e-7,0.8361852,0.1560841,0.003131095,0.002678369,0.001286348,0.0001042665,0.00009659825,0.000433844],"genre_scores_gemma":[4.081618e-7,0.6221713,0.3725325,0.000233924,0.0008220475,0.0003002607,0.002319304,0.00004466142,0.001575534],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9546065,"threshold_uncertainty_score":0.9999236,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4889105727292296,"score_gpt":0.6829451091120035,"score_spread":0.1940345363827738,"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."}}