{"id":"W4400066081","doi":"10.14245/ns.2448060.030","title":"The Quantitative Evaluation of Automatic Segmentation in Lumbar Magnetic Resonance Images","year":2024,"lang":"en","type":"article","venue":"Neurospine","topic":"Medical Imaging and Analysis","field":"Engineering","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Food and Drug Administration","keywords":"Segmentation; Magnetic resonance imaging; Computer science; Lumbar; Artificial intelligence; Spinal stenosis; Lumbar spinal stenosis; Neurogenic claudication; Residual; Radiology; Medicine; Computer vision; Algorithm","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":[0.005941466,0.001234212,0.0009782307,0.004735758,0.0005170319,0.002428457,0.001102862,0.001995621,0.001268132],"category_scores_gemma":[0.01689819,0.0003969712,0.0007292743,0.001650631,0.001124937,0.00150973,0.001159534,0.0004404373,0.0008003032],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001204087,"about_ca_system_score_gemma":0.001107057,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003171837,"about_ca_topic_score_gemma":0.002961139,"domain_scores_codex":[0.9951687,0.001149974,0.0004483213,0.0008649807,0.002139443,0.000228629],"domain_scores_gemma":[0.9922058,0.003503678,0.001462485,0.0006670298,0.002016512,0.000144421],"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":[0.002151778,0.0004230589,0.06373487,0.001603418,0.0008163598,0.0004083585,0.0005378316,0.2774864,0.08987749,0.002717647,0.003937731,0.5563051],"study_design_scores_gemma":[0.00004079814,0.0006660799,0.04022307,0.00015657,0.000144348,0.0008250513,0.0001642281,0.8984408,0.05446703,0.001926261,0.0028695,0.00007625898],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5304425,0.00469033,0.4510144,0.0004034663,0.0001752031,0.0004542381,0.001621071,0.005551402,0.005647419],"genre_scores_gemma":[0.8728803,0.0007649669,0.1221693,0.0001353342,0.00008146231,0.0001455028,0.002243324,0.0003196097,0.001260183],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005941466,"threshold_uncertainty_score":0.03142184,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01788857608803217,"score_gpt":0.3015342366710155,"score_spread":0.2836456605829833,"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."}}