{"id":"W4413324439","doi":"10.1016/j.cpet.2025.07.006","title":"Objective Task-Based Evaluation of Quantitative Medical Imaging Methods","year":2025,"lang":"en","type":"review","venue":"PET Clinics","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Medicine; Task (project management); Medical physics; Systems engineering","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"],"consensus_categories":[],"category_scores_codex":[0.01616455,0.000398986,0.00284762,0.0005259675,0.00006606505,0.00001773435,0.0003328793,0.000263297,0.0003290699],"category_scores_gemma":[0.06384675,0.0003130703,0.0009277957,0.0006868789,0.0003251689,0.00003754732,0.0001201425,0.001846449,0.00001736176],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002552653,"about_ca_system_score_gemma":0.007757006,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000292487,"about_ca_topic_score_gemma":0.000002140699,"domain_scores_codex":[0.9931366,0.00278623,0.001598627,0.0006262892,0.00154929,0.0003029749],"domain_scores_gemma":[0.988504,0.008797234,0.00100551,0.0006064682,0.0008338118,0.0002529855],"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":[0.00001921982,0.0001459652,0.00004938176,0.01532814,0.0003251856,0.00004461697,0.00005426597,0.0000126447,2.211645e-7,0.0003369913,0.0006150614,0.9830683],"study_design_scores_gemma":[0.00187024,0.0001222063,0.00002745952,0.06028114,0.01068421,0.00009475673,0.00007375051,0.2532707,9.972671e-7,0.0006569839,0.6725907,0.0003268881],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000004177037,0.9417537,0.05127295,0.0005529076,0.001072665,0.001049866,0.00002956779,0.0000587637,0.004205446],"genre_scores_gemma":[0.00001365737,0.9297327,0.06879057,0.0006232096,0.0002038801,0.0001086648,0.0003201332,0.0000585831,0.000148644],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9827414,"threshold_uncertainty_score":0.9999321,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08863096550514929,"score_gpt":0.577972175259264,"score_spread":0.4893412097541148,"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."}}