{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01452248,0.002308267,0.001567501,0.006082905,0.0009745804,0.004044905,0.002372714,0.0017048,0.009748331],"category_scores_gemma":[0.026009,0.0006443295,0.001974676,0.003205467,0.0009663113,0.00250499,0.003626605,0.001685427,0.003647368],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001456011,"about_ca_system_score_gemma":0.002062016,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001603561,"about_ca_topic_score_gemma":0.001426835,"domain_scores_codex":[0.9863288,0.004990411,0.001448522,0.001135674,0.005650941,0.0004454969],"domain_scores_gemma":[0.9846938,0.006171506,0.00129258,0.0020533,0.005239835,0.000549097],"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":[0.001252377,0.001081684,0.005088668,0.001704232,0.0005909635,0.0008507157,0.001240272,0.2494664,0.05053414,0.07511643,0.02528702,0.5877871],"study_design_scores_gemma":[0.0001239516,0.0007692711,0.00331839,0.0002589808,0.0001066226,0.0003702563,0.0002255768,0.9064322,0.02677378,0.03553186,0.02593695,0.000152202],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.007480138,0.0001553133,0.9838334,0.000132107,0.00004476302,0.0007381442,0.00047885,0.003966777,0.003170502],"genre_scores_gemma":[0.07323831,0.0001328162,0.9204257,0.00007088877,0.00004732462,0.00149356,0.001530935,0.0009036271,0.002156845],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01452248,"threshold_uncertainty_score":0.07680315,"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."}}