{"id":"W4413846771","doi":"10.1016/j.ijcha.2025.101783","title":"Evaluating artificial intelligence-enabled medical tests in cardiology: Best practice","year":2025,"lang":"en","type":"review","venue":"IJC Heart & Vasculature","topic":"Cardiac Imaging and Diagnostics","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal; Montreal Heart Institute","funders":"Novo Nordisk Fonden; Hartstichting; Austrian Science Fund; ZonMw; Villum Fonden; Novo Nordisk; Danish Cardiovascular Academy; European Federation of Pharmaceutical Industries and Associations; HORIZON EUROPE Framework Programme; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; National Institutes of Health; Hjerteforeningen; European Commission; Deutsche Forschungsgemeinschaft; Danish Data Science Academy","keywords":"Medicine; Medical physics; Cardiology; Internal medicine","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.003505373,0.0006093791,0.003673231,0.0005105448,0.0001097165,0.00007345665,0.0002343868,0.001652812,0.00009567232],"category_scores_gemma":[0.05566695,0.000496384,0.001482042,0.001292025,0.000156877,0.00008967507,0.0002214201,0.003543192,0.0003601846],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003219533,"about_ca_system_score_gemma":0.005010497,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001263815,"about_ca_topic_score_gemma":0.00001765118,"domain_scores_codex":[0.9948327,0.001173017,0.001176724,0.0009299791,0.001194938,0.0006926241],"domain_scores_gemma":[0.9917974,0.006189541,0.0001983673,0.001029519,0.0004235319,0.0003615904],"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":[0.00002844766,0.0002507516,0.0001480229,0.01339204,0.0005571694,0.001390742,0.00006983026,0.00004633603,4.706901e-7,0.0001511665,0.008696601,0.9752684],"study_design_scores_gemma":[0.0002352134,0.0001364267,0.0001013587,0.05409633,0.004805088,0.002177333,0.00008091983,0.00006407004,7.232829e-7,0.0001268545,0.937784,0.0003916744],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001929476,0.9882348,0.0001114726,0.002107165,0.001640462,0.00158729,0.00002643047,0.00010912,0.006163998],"genre_scores_gemma":[0.00006782273,0.9931225,0.002303066,0.001164695,0.002432629,0.0002023471,0.0002837198,0.00007234987,0.0003509264],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9748768,"threshold_uncertainty_score":0.9997488,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0982604628301758,"score_gpt":0.4782843423211831,"score_spread":0.3800238794910072,"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."}}