{"id":"W7149001746","doi":"10.2196/87163","title":"Symptom-Only Localization of Brainstem Ischemia: Large Language Models vs. Neurologists in 109 Diffusion-Weighted Imaging–Positive Cases: A Retrospective Study (Preprint)","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"Retrospective cohort study; Brainstem; Language model; Neuroimaging; MEDLINE","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001635589,0.0006856253,0.0005490967,0.001987428,0.0009448094,0.001475391,0.0007657208,0.0009243921,0.004829962],"category_scores_gemma":[0.01082905,0.0007168193,0.0009084473,0.001034643,0.001123037,0.002201777,0.0008910007,0.0006931551,0.001582139],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005369354,"about_ca_system_score_gemma":0.0006141078,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00414269,"about_ca_topic_score_gemma":0.003622109,"domain_scores_codex":[0.9983503,0.0004997161,0.0002333866,0.0004411515,0.0001851764,0.0002903651],"domain_scores_gemma":[0.9957112,0.002023869,0.001018967,0.0004676799,0.0003795501,0.0003986815],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004168561,0.000134062,0.9949497,0.00001989387,0.00005475289,0.002408594,0.0004341155,0.00003928755,0.0002730766,0.00003552578,0.00009145839,0.001142699],"study_design_scores_gemma":[0.00007517274,0.001057988,0.9689524,0.0000772925,0.0004598149,0.01818445,0.007957381,0.001408877,0.0007827796,0.0003601323,0.0006239982,0.0000598338],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9991788,0.0001026359,0.0001232417,0.0000341564,0.000005338411,0.00001232972,0.0001305633,0.000002520571,0.0004103757],"genre_scores_gemma":[0.9995147,0.00007804504,0.0000575681,0.00002567193,0.0000160916,0.000008297435,0.0002053336,0.000007283758,0.00008715477],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004829962,"threshold_uncertainty_score":0.01615787,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01537062604842657,"score_gpt":0.3603713598411399,"score_spread":0.3450007337927133,"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."}}