{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005198474,0.0001025771,0.0001511424,0.0001726296,0.0001859252,0.00001512823,0.0004020593,0.0000341677,0.004683428],"category_scores_gemma":[0.00003639562,0.0001148088,0.00004437206,0.0002506752,0.0001216725,0.0006323214,0.0002478772,0.0002198473,6.970661e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004531625,"about_ca_system_score_gemma":0.0004502654,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007488682,"about_ca_topic_score_gemma":9.698829e-7,"domain_scores_codex":[0.9986265,0.0002847074,0.0001222735,0.0003401356,0.0005001432,0.0001261928],"domain_scores_gemma":[0.9993266,0.0000888137,0.000114234,0.0002768459,0.00006196569,0.0001315758],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001540383,0.00301884,0.6590247,0.00008768665,0.0005398261,0.0001088919,0.006196602,0.01254734,0.003587886,0.05020266,0.06767108,0.1954741],"study_design_scores_gemma":[0.007101154,0.0007440665,0.01780044,0.0002033256,0.0001857279,0.00001239231,0.007301454,0.2659141,0.01290075,0.003240325,0.6837214,0.0008748922],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.878545,0.0004681993,0.08294353,0.006554969,0.0004058552,0.000615015,0.0001446935,0.0002827739,0.03003996],"genre_scores_gemma":[0.9819552,0.00001501086,0.01446652,0.0001189358,0.0001926447,0.000006456671,0.0001991339,0.00002250857,0.003023636],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6412242,"threshold_uncertainty_score":0.9962264,"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."}}