{"id":"W3092461455","doi":"10.1016/j.nic.2020.08.004","title":"Machine Learning Algorithm Validation","year":2020,"lang":"en","type":"review","venue":"Neuroimaging Clinics of North America","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":132,"is_retracted":false,"has_abstract":true,"ca_institutions":"Royal Victoria Hospital; Jewish General Hospital; McGill University; Royal Victoria Regional Health Centre; University of Saskatchewan; McGill University Health Centre","funders":"Fonds de Recherche du Québec - Santé; Fondation de l'Association des radiologistes du Québec","keywords":"Software deployment; Workflow; Domain (mathematical analysis); Model validation; Medicine; Machine learning; Health care; Artificial intelligence; Computer science; Medical physics; Algorithm; Data science; Software engineering","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":[0.03036205,0.001842723,0.002833595,0.004597746,0.0008197086,0.003878888,0.00342994,0.002294797,0.01199796],"category_scores_gemma":[0.08946546,0.000574101,0.0027326,0.003418593,0.00127008,0.002044625,0.001864079,0.003597787,0.006742843],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001761679,"about_ca_system_score_gemma":0.006446111,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003286104,"about_ca_topic_score_gemma":0.002870022,"domain_scores_codex":[0.9834945,0.008273224,0.001881399,0.001830132,0.004187468,0.0003333051],"domain_scores_gemma":[0.9582906,0.02743345,0.001728447,0.003497192,0.008751018,0.0002992701],"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":[0.0002282905,0.0002097259,0.004011416,0.0076416,0.001298139,0.0001327876,0.0001443198,0.03483226,0.0009392979,0.02869655,0.03704948,0.8848162],"study_design_scores_gemma":[0.0002909385,0.0007465883,0.008277752,0.01807541,0.001184936,0.001149057,0.00028043,0.2305627,0.01002836,0.08686553,0.6423097,0.0002286689],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.01026385,0.1688335,0.76096,0.005586128,0.003058786,0.002319414,0.005023157,0.003871303,0.04008381],"genre_scores_gemma":[0.162196,0.1198548,0.671898,0.004126224,0.001897188,0.00422103,0.01960227,0.001904046,0.01430054],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.9696379,"threshold_uncertainty_score":0.1605718,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2090802705330677,"score_gpt":0.4713114398952047,"score_spread":0.262231169362137,"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."}}