{"id":"W2536315965","doi":"10.1109/embc.2016.7592203","title":"A multi-criteria evaluation platform for segmentation algorithms","year":2016,"lang":"en","type":"article","venue":"","topic":"Medical Image Segmentation Techniques","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre Hospitalier de l’Université de Montréal","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Segmentation; Computer science; Robustness (evolution); Ground truth; Algorithm; Image segmentation; Reliability (semiconductor); Data mining; Artificial intelligence; Outlier; Sensitivity (control systems); Graph; Machine learning; Theoretical computer science","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":[0.0007998779,0.00008268626,0.00007678678,0.00008272713,0.00005966709,0.00008217498,0.0003027014,0.00003988553,0.0003264073],"category_scores_gemma":[0.0002085846,0.00005249868,0.00003761628,0.0001096897,0.00002704348,0.001117931,0.00006130677,0.00001909825,0.00005196028],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000107744,"about_ca_system_score_gemma":0.00005088445,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008025204,"about_ca_topic_score_gemma":0.000003748914,"domain_scores_codex":[0.9989458,0.00003596038,0.0002207709,0.000269266,0.000366259,0.0001619375],"domain_scores_gemma":[0.999223,0.000131291,0.00007278804,0.0002471882,0.0002516762,0.00007404501],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.000002473136,0.00003029961,0.000008839896,0.000004501652,0.00000490479,2.333791e-7,0.0001722843,1.296077e-7,0.1542224,0.0004716121,0.003882286,0.8412],"study_design_scores_gemma":[0.002169064,0.0001361328,0.0003201606,0.00002601081,0.000008760639,0.000003787042,0.00004170786,0.2390258,0.7536597,0.003960937,0.0004870839,0.0001608428],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0004552907,0.0000110957,0.9970233,0.0008311843,0.0002748402,0.0008380637,0.000004671859,0.0003384968,0.0002230447],"genre_scores_gemma":[0.01374125,0.000008125762,0.9843386,0.0006438246,0.00004639474,0.0004479629,0.00001083376,0.00000653131,0.0007564453],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8410392,"threshold_uncertainty_score":0.3573931,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08988778626439302,"score_gpt":0.3948647655187402,"score_spread":0.3049769792543471,"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."}}