{"id":"W4295249529","doi":"10.1088/1361-6560/ac8044","title":"OpenKBP-Opt: an international and reproducible evaluation of 76 knowledge-based planning pipelines","year":2022,"lang":"en","type":"article","venue":"YUHSpace (Yonsei University Medical Library)","topic":"Advanced Radiotherapy Techniques","field":"Physics and Astronomy","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"Princess Margaret Cancer Centre; Vector Institute; University of Toronto","funders":"National Cancer Institute","keywords":"Wilcoxon signed-rank test; Radiation treatment planning; Reference dose; Voxel; Computer science; Test plan; Nuclear medicine; Pipeline (software); Contouring; Set (abstract data type); Mathematics; Radiation therapy; Medicine; Statistics; Artificial intelligence; Surgery; Weibull distribution","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02250772,0.001854884,0.001092,0.002775754,0.001011537,0.002499973,0.003873293,0.001801726,0.003779557],"category_scores_gemma":[0.0450093,0.0009954172,0.001960865,0.002234982,0.001496137,0.002882295,0.005555017,0.001717012,0.0009277787],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004410693,"about_ca_system_score_gemma":0.005456313,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01535938,"about_ca_topic_score_gemma":0.01692862,"domain_scores_codex":[0.983017,0.008547516,0.001226045,0.001990495,0.004588575,0.0006303703],"domain_scores_gemma":[0.9731134,0.01349179,0.001478439,0.006698424,0.004325556,0.0008924233],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"observational","study_design_scores_codex":[0.00314209,0.002907842,0.02247719,0.00209483,0.001014074,0.0003374293,0.00107747,0.5172439,0.01007456,0.005352636,0.02023427,0.4140438],"study_design_scores_gemma":[0.001230483,0.002424247,0.02520094,0.0002979964,0.0003204735,0.0003279019,0.0006902303,0.9140915,0.02663516,0.005354924,0.0232369,0.0001891808],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5146275,0.002226952,0.4004177,0.001022098,0.0003728894,0.005797943,0.02228655,0.03030429,0.02294407],"genre_scores_gemma":[0.584592,0.0003039528,0.383445,0.0002220962,0.00003778165,0.002599031,0.02447529,0.002500546,0.001824197],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9774923,"threshold_uncertainty_score":0.1190336,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02594673138371828,"score_gpt":0.3087785677969491,"score_spread":0.2828318364132308,"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."}}