{"id":"W4377004088","doi":"10.59275/j.melba.2022-2d93","title":"Rethinking Generalization: The Impact of Annotation Style on Medical Image Segmentation","year":2022,"lang":"en","type":"article","venue":"The Journal of Machine Learning for Biomedical Imaging","topic":"Radiomics and Machine Learning in Medical Imaging","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"NeuroRx Research (Canada); McGill University; Mila - Quebec Artificial Intelligence Institute","funders":"Genentech; MedDay Pharmaceuticals; Royal Academy of Engineering; International Progressive MS Alliance; Compute Canada; Western Canada Research Grid; Canadian Institute for Advanced Research; Teva Pharmaceutical Industries; Biogen","keywords":"Annotation; Computer science; Context (archaeology); Segmentation; Generalization; Artificial intelligence; Machine learning; Style (visual arts); Ground truth; Automatic image annotation; Image (mathematics); Natural language processing; Image retrieval; Information retrieval; Mathematics; Geography","routes":{"ca_aff":true,"ca_fund":true,"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":[],"consensus_categories":[],"category_scores_codex":[0.007355106,0.0001821933,0.0003908626,0.0002954957,0.0007183448,0.00003656709,0.0004554645,0.00004273121,0.0005456441],"category_scores_gemma":[0.003429352,0.00009591263,0.0003541499,0.0005020382,0.0003831286,0.0001129966,0.0001373455,0.001905832,0.000001425909],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002167974,"about_ca_system_score_gemma":0.0003425198,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000154434,"about_ca_topic_score_gemma":5.09462e-7,"domain_scores_codex":[0.9961895,0.0008263643,0.0008312141,0.0001501221,0.001710976,0.0002918118],"domain_scores_gemma":[0.997189,0.001178773,0.0009568654,0.0002049443,0.0002657615,0.0002045889],"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":[0.007996956,0.002219351,0.06173733,0.0005498881,0.002116714,0.0004629012,0.03370019,0.1421036,0.1403581,0.001380096,0.05328485,0.55409],"study_design_scores_gemma":[0.004827214,0.002394076,0.006394972,0.0002587567,0.0003599918,0.002648264,0.001076035,0.9730359,0.0004662011,0.001163022,0.007201214,0.0001743977],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7485507,0.001444598,0.1844354,0.063624,0.0008104965,0.0006674146,0.00001789067,0.00005863958,0.0003909035],"genre_scores_gemma":[0.994755,0.0001283996,0.00278199,0.001471822,0.0005915483,0.0000101675,0.00009895772,0.00004563849,0.0001164931],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8309323,"threshold_uncertainty_score":0.8279997,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01342492877643922,"score_gpt":0.3338452928445346,"score_spread":0.3204203640680953,"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."}}