{"id":"W4376473362","doi":"10.1615/critrevoncog.2023048797","title":"Perspective of a Pathologist on Benchmark Strategies for Artificial Intelligence Development in Organ Transplantation","year":2023,"lang":"en","type":"review","venue":"Critical Reviews™ in Oncogenesis","topic":"Prenatal Screening and Diagnostics","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"University Hospital Foundation","funders":"","keywords":"Interpretability; Digital pathology; Transplantation; Medicine; Process (computing); Computer science; Gold standard (test); Intensive care medicine; Artificial intelligence; Medical physics; Pathology; Surgery; Radiology","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":[0.00785403,0.0008297738,0.0009273369,0.002623851,0.0005854947,0.003610031,0.001955144,0.004552708,0.005322212],"category_scores_gemma":[0.008667124,0.0002971468,0.0007855663,0.001966071,0.002206223,0.005102727,0.002271567,0.005073534,0.003352768],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002322582,"about_ca_system_score_gemma":0.004653338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001620453,"about_ca_topic_score_gemma":0.002527344,"domain_scores_codex":[0.9983134,0.0007746972,0.0001900357,0.0001474581,0.0004338837,0.0001405202],"domain_scores_gemma":[0.9940567,0.003263912,0.0002929386,0.0001894714,0.001719921,0.0004770807],"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":[0.0001170006,0.0001101843,0.0003266535,0.0114737,0.00006218098,0.0005769396,0.0005273285,0.0009290789,0.001032806,0.06665855,0.141253,0.7769327],"study_design_scores_gemma":[0.00001884439,0.000112007,0.0003299767,0.008660633,0.00003023538,0.0007294156,0.0003003611,0.0001486368,0.0003488674,0.01713633,0.9721624,0.00002220202],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002900953,0.9534599,0.001485146,0.03383909,0.003048967,0.00002443974,0.00003760489,0.00003586655,0.007778903],"genre_scores_gemma":[0.00659184,0.9637398,0.004206414,0.01759446,0.00309235,0.00009812343,0.00008760016,0.00002865249,0.004560777],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.992146,"threshold_uncertainty_score":0.04153657,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2538991459600352,"score_gpt":0.478408049979496,"score_spread":0.2245089040194608,"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."}}