{"id":"W4390992147","doi":"10.1109/bibm58861.2023.10385410","title":"A Closer Look at Weak Supervision’s Limitations in WSI Recurrence Score Prediction","year":2023,"lang":"en","type":"article","venue":"","topic":"AI in cancer detection","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University; University of Alberta","funders":"","keywords":"Computer science; Notice; Cluster analysis; Artificial intelligence; Graphical user interface; Machine learning; F1 score; Calculator; Pattern recognition (psychology)","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":[],"consensus_categories":[],"category_scores_codex":[0.02461944,0.001659907,0.001503818,0.001447908,0.0007854485,0.002322655,0.00260754,0.001769611,0.001459463],"category_scores_gemma":[0.05552329,0.0004607594,0.0008923316,0.001082533,0.001255345,0.002973805,0.001986536,0.003337229,0.0007885995],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008410292,"about_ca_system_score_gemma":0.0009298584,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009840472,"about_ca_topic_score_gemma":0.01146384,"domain_scores_codex":[0.9918783,0.004685546,0.0004115508,0.001724433,0.001047206,0.0002528529],"domain_scores_gemma":[0.9594895,0.03161921,0.001251995,0.00379369,0.003134235,0.0007113845],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009471346,0.0003619605,0.1473939,0.0007905124,0.0009031616,0.0004898223,0.0007584381,0.155302,0.007835736,0.007216767,0.02829535,0.6497052],"study_design_scores_gemma":[0.00002131224,0.0002145926,0.01428407,0.0001010979,0.00008846104,0.0001465587,0.0001191777,0.9687828,0.003967386,0.009104043,0.003132549,0.00003798702],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3693305,0.02557387,0.5672362,0.02067259,0.0008589471,0.0001915717,0.002192545,0.005191531,0.00875222],"genre_scores_gemma":[0.92897,0.001586128,0.06301687,0.0009767,0.0005965032,0.00006766635,0.002413779,0.0003101511,0.002062214],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02461944,"threshold_uncertainty_score":0.1302016,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08246528623866789,"score_gpt":0.2705518893429427,"score_spread":0.1880866031042748,"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."}}