{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07717753,0.003688865,0.009079892,0.01519915,0.00128703,0.009817036,0.01086012,0.006507037,0.0103722],"category_scores_gemma":[0.2003202,0.00220177,0.007292978,0.01041814,0.003745065,0.01015755,0.005560802,0.009274583,0.00667708],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006014476,"about_ca_system_score_gemma":0.03192437,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01683583,"about_ca_topic_score_gemma":0.02248092,"domain_scores_codex":[0.9673914,0.01192615,0.009369601,0.002039926,0.008583888,0.0006890644],"domain_scores_gemma":[0.8201622,0.1012427,0.01337104,0.00847346,0.05459663,0.002153843],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002681373,0.0001568229,0.001209193,0.06278482,0.0009811234,0.0001288407,0.0002646019,0.001137233,0.000693782,0.01513769,0.1411625,0.7760753],"study_design_scores_gemma":[0.0003842352,0.0003102834,0.003800667,0.1947617,0.003921177,0.0004430915,0.000393274,0.003166276,0.00180743,0.04131475,0.74941,0.0002871141],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0004885616,0.9462669,0.02067477,0.02032833,0.005011596,0.0009589425,0.001954024,0.0008729456,0.003443924],"genre_scores_gemma":[0.006691962,0.8197082,0.1446941,0.01490213,0.002395995,0.002775452,0.005677405,0.0005639837,0.002590828],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9228225,"threshold_uncertainty_score":0.4081587,"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."}}