{"id":"W4402474868","doi":"10.1109/ccece59415.2024.10667319","title":"Assessing the Performance of Foundation Models in Prostate Segmentation Across Different Ultrasound Modalities","year":2024,"lang":"en","type":"article","venue":"","topic":"Medical Image Segmentation Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; Queen's University","funders":"","keywords":"Modalities; Foundation (evidence); Computer science; Segmentation; Artificial intelligence; Image segmentation; Geography","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":[0.002740379,0.002531641,0.001289128,0.001905291,0.0006808557,0.001898718,0.001730822,0.00254309,0.001622102],"category_scores_gemma":[0.006369287,0.0007244732,0.002692166,0.001004201,0.0007060183,0.001272621,0.001563617,0.001899776,0.00129508],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001205865,"about_ca_system_score_gemma":0.001677567,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02085928,"about_ca_topic_score_gemma":0.02901743,"domain_scores_codex":[0.9990221,0.0002195153,0.00007319282,0.0003391254,0.0002044524,0.0001415],"domain_scores_gemma":[0.9979872,0.001155036,0.0001347411,0.0002825393,0.0003084379,0.0001320048],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001189546,0.0004091516,0.01815268,0.0007588325,0.0007766807,0.000268135,0.0001640514,0.7229136,0.01200658,0.00140785,0.01359475,0.2283581],"study_design_scores_gemma":[0.00004127523,0.0001970245,0.002741178,0.00007470985,0.00007991665,0.0001606275,0.00005975773,0.9870245,0.005671021,0.0008815603,0.003029905,0.00003851099],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5838689,0.0166852,0.3337013,0.003255975,0.00154999,0.0006426876,0.01286996,0.03610109,0.01132496],"genre_scores_gemma":[0.8088867,0.002330397,0.159475,0.001119922,0.000184242,0.000194261,0.02246293,0.001378624,0.003967984],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02085928,"threshold_uncertainty_score":0.04147571,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03479306637790007,"score_gpt":0.3544605530889108,"score_spread":0.3196674867110107,"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."}}