2f:I[703058,["/_next/static/immutable/chunks/3bhq_5eh1d760.js","/_next/static/immutable/chunks/1855-ogyi8eh4.js","/_next/static/immutable/chunks/2pfsjpn6k6zlg.js","/_next/static/immutable/chunks/26c7v3_3uwnwu.js","/_next/static/immutable/chunks/25a3ozcrvi1ff.js","/_next/static/immutable/chunks/2r4wx50nvjh_4.js","/_next/static/immutable/chunks/23fsw06hsi5b1.js","/_next/static/immutable/chunks/06_yt270h41rg.js","/_next/static/immutable/chunks/3ezsiz-7shkna.js","/_next/static/immutable/chunks/20x2y96ldu5wx.js","/_next/static/immutable/chunks/2lxntcm3pd1_h.js","/_next/static/immutable/chunks/00dgco8irq3jw.js","/_next/static/immutable/chunks/00nwe-kq0rboc.js","/_next/static/immutable/chunks/2051ffx3xzio9.js"],"PaperListPageClient"] 30:T770,Study DesignDiagnostic accuracy study.ObjectivePrior evaluations of frontier models as radiology decision-support tools relied on 2-dimensional images or text reports; their ability to interpret volumetric data remains unclear. This study assessed Google Gemini 3 Pro for grading lumbar spinal canal stenosis on video lumbar magnetic resonance imaging (MRI) and evaluated diagnostic accuracy, agreement with neuroradiologist consensus, and the effect of localizer-assisted input.MethodsThe Radiological Society of North America (RSNA) 2024 Lumbar Spine Degenerative Classification Dataset, with American Society of Neuroradiology (ASNR) consensus labels, served as a reference benchmark; interobserver agreement among contributing readers was not reported. 100 examinations yielded 500 disc-level observations (371 normal/mild, 74 moderate, 55 severe), demonstrating marked class imbalance. Native imaging series were converted into synchronized video montages. Gemini 3 Pro generated one grade per disc level with and without localizer overlays. Primary outcome was linearly weighted kappa (κw); secondary outcomes included class-wise performance, severe-case error patterns, and overall accuracy.ResultsWithout localizer, overall accuracy was 75.6% (378/500) with fair agreement (κw = 0.39). Severe stenosis sensitivity was 41.8%; 43.6% of severe cases were downgraded to normal/mild, and 58.2% to non-severe. With localizer overlays, accuracy was 73.2% (366/500) with κw = 0.32, and severe sensitivity decreased to 30.9%; severe-to-normal/mild misses increased to 52.7%. Differences were not significant.ConclusionsGemini 3 Pro showed fair agreement with the neuroradiologist consensus benchmark, but apparent overall accuracy was inflated by the majority normal/mild class and masked clinically unacceptable under-detection of severe stenosis. Localizer-assisted input did not improve performance.31:Ta28, Study Design Diagnostic accuracy study. Objective Prior evaluations of frontier models as radiology decision-support tools relied on 2-dimensional images or text reports; their ability to interpret volumetric data remains unclear. This study assessed Google Gemini 3 Pro for grading lumbar spinal canal stenosis on video lumbar magnetic resonance imaging (MRI) and evaluated diagnostic accuracy, agreement with neuroradiologist consensus, and the effect of localizer-assisted input. Methods The Radiological Society of North America (RSNA) 2024 Lumbar Spine Degenerative Classification Dataset, with American Society of Neuroradiology (ASNR) consensus labels, served as a reference benchmark; interobserver agreement among contributing readers was not reported. 100 examinations yielded 500 disc-level observations (371 normal/mild, 74 moderate, 55 severe), demonstrating marked class imbalance. Native imaging series were converted into synchronized video montages. Gemini 3 Pro generated one grade per disc level with and without localizer overlays. Primary outcome was linearly weighted kappa (κw); secondary outcomes included class-wise performance, severe-case error patterns, and overall accuracy. Results Without localizer, overall accuracy was 75.6% (378/500) with fair agreement (κw = 0.39). Severe stenosis sensitivity was 41.8%; 43.6% of severe cases were downgraded to normal/mild, and 58.2% to non-severe. With localizer overlays, accuracy was 73.2% (366/500) with κw = 0.32, and severe sensitivity decreased to 30.9%; severe-to-normal/mild misses increased to 52.7%. Differences were not significant. Conclusions Gemini 3 Pro showed fair agreement with the neuroradiologist consensus benchmark, but apparent overall accuracy was inflated by the majority normal/mild class and masked clinically unacceptable under-detection of severe stenosis. Localizer-assisted input did not improve performance. 25:[[["$","script","0",{"type":"application/ld+json","dangerouslySetInnerHTML":{"__html":"{\"@context\":\"https://schema.org\",\"@type\":\"CollectionPage\",\"name\":\"Multimodal AI / Large Language Model: modality Research & Clinical Takeaways\",\"description\":\"Explore peer-reviewed papers on Multimodal AI / Large Language Model (modality). Clinician-written summaries and actionable practice insights. Updated daily.\",\"url\":\"https://www.reachrx.ai/hub/tag/modality/Multimodal%20AI%20%2F%20Large%20Language%20Model\",\"breadcrumb\":{\"@type\":\"BreadcrumbList\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https://www.reachrx.ai\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Research Hub\",\"item\":\"https://www.reachrx.ai/hub\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Modalities\",\"item\":\"https://www.reachrx.ai/hub/tag/modality\"},{\"@type\":\"ListItem\",\"position\":4,\"name\":\"Multimodal AI / Large Language Model\",\"item\":\"https://www.reachrx.ai/hub/tag/modality/Multimodal%20AI%20%2F%20Large%20Language%20Model\"}]}}"}}]],["$","$L2f",null,{"title":"Multimodal AI / Large Language Model: Research & Clinical Takeaways","description":"Review evidence by treatment modality: Multimodal AI / Large Language Model — concise summaries and practical next steps.","papers":[{"doi":"10.1177/21925682261448823","pmid":"42092340","title":"Benchmarking Multimodal Vision Frontier Models With Lumbar Spine MRIs for Grading Lumbar Spinal Stenosis","journal_title":"Global Spine Journal","journal_issn":"2192-5682","published_at":"2026-05-07T00:00:00.000Z","url":"https://doi.org/10.1177/21925682261448823","source_score":38,"is_retracted":false,"is_open_access":true,"raw_metadata":{"ooir":{"day":"2026-05-11","doi":"10.1177/21925682261448823","days":"7","issn":"2192-5682","rank":"8","field":"Clinical Medicine","score":"38","category":"Clinical Neurology","categories":["Clinical Neurology"],"_categories":["Clinical Neurology"]},"pubmed":{"doi":"10.1177/21925682261448823","title":"Benchmarking Multimodal Vision Frontier Models With Lumbar Spine MRIs for Grading Lumbar Spinal Stenosis.","journal":"Global spine journal","methods":"","results":"","abstract":"$30","keywords":["MRI cine video","diagnostic accuracy","inter-rater reliability","lumbar spinal canal stenosis","multimodal vision frontier models"],"pubmed_id":"42092340","conclusions":"","publication_date":"2026-05-07T00:00:00"},"crossref":{"DOI":"10.1177/21925682261448823","URL":"https://doi.org/10.1177/21925682261448823","ISSN":["2192-5682","2192-5690"],"link":[{"URL":"https://journals.sagepub.com/doi/pdf/10.1177/21925682261448823","content-type":"application/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https://journals.sagepub.com/doi/full-xml/10.1177/21925682261448823","content-type":"application/xml","content-version":"vor","intended-application":"text-mining"},{"URL":"https://journals.sagepub.com/doi/pdf/10.1177/21925682261448823","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"page":null,"type":"journal-article","issue":null,"score":1,"title":["Benchmarking Multimodal Vision Frontier Models With Lumbar Spine MRIs for Grading Lumbar Spinal Stenosis"],"author":[{"given":"Harry","family":"Gebhard","sequence":"first","affiliation":[{"name":"Medical University Lausitz – Carl Thiem"}]},{"given":"Ahmet","family":"Kartal","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Noel F.","family":"Manalil","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Lawrance K.","family":"Chung","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Sohail R.","family":"Daulat","sequence":"additional","affiliation":[{"name":"Department of Neurosurgery, University of Arizona College of Medicine–Tucson"}]},{"given":"Chiungwen D.","family":"Cheng","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Ayesha Akbar","family":"Waheed","sequence":"additional","affiliation":[{"name":"University of Pittsburgh School of Medicine"}]},{"given":"Andia","family":"Shahzadi","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Ibrahim","family":"Hussain","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Roger","family":"Härtl","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]},{"given":"Galal A.","family":"Elsayed","sequence":"additional","affiliation":[{"name":"Och Spine at NewYork-Presbyterian Hospital, Weill Cornell Medicine"}]}],"issued":{"date-parts":[[2026,5,6]]},"member":"179","prefix":"10.1177","source":"Crossref","volume":null,"created":{"date-time":"2026-05-07T03:59:22Z","timestamp":1778126362000,"date-parts":[[2026,5,7]]},"indexed":{"date-time":"2026-05-10T08:08:22Z","timestamp":1778400502840,"date-parts":[[2026,5,10]]},"license":[{"URL":"https://creativecommons.org/licenses/by-nc-nd/4.0/","start":{"date-time":"2026-05-06T00:00:00Z","timestamp":1778025600000,"date-parts":[[2026,5,6]]},"delay-in-days":0,"content-version":"vor"},{"URL":"https://journals.sagepub.com/page/policies/text-and-data-mining-license","start":{"date-time":"2026-05-06T00:00:00Z","timestamp":1778025600000,"date-parts":[[2026,5,6]]},"delay-in-days":0,"content-version":"tdm"}],"subject":[],"abstract":"$31","language":"en","relation":{},"resource":{"primary":{"URL":"https://journals.sagepub.com/doi/10.1177/21925682261448823"}},"subtitle":[],"deposited":{"date-time":"2026-05-10T07:17:38Z","timestamp":1778397458000,"date-parts":[[2026,5,10]]},"issn-type":[{"type":"print","value":"2192-5682"},{"type":"electronic","value":"2192-5690"}],"published":{"date-parts":[[2026,5,6]]},"publisher":"SAGE Publications","reference":[{"DOI":"10.48550/arXiv.2311.05591","key":"e_1_3_7_2_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.2196/67409","key":"e_1_3_7_3_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":"Sorich MJ Mangoni AA Bacchi S Menz BD Hopkins AM. The triage and diagnostic accuracy of frontier large language models: updated comparison to physician performance. J Med Internet Res. 2024;26:e67409. doi:10.2196/67409. (In eng).","article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.2196/59505","key":"e_1_3_7_4_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.48550/arXiv.2512.13747","key":"e_1_3_7_5_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":"Dai S Li L Zhao K et al. Why Text Prevails: Vision May Undermine Multimodal Medical Decision Making. arXiv preprint arXiv:251213747. doi:10.48550/arXiv.2512.13747.","article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1007/s00256-025-05027-9","key":"e_1_3_7_6_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3390/diagnostics15070930","key":"e_1_3_7_7_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3348/kjr.2025.0599","key":"e_1_3_7_8_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1038/s41746-025-01649-4","key":"e_1_3_7_9_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3390/tomography9050145","key":"e_1_3_7_10_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1590/0103-6440202305499","key":"e_1_3_7_11_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":null,"key":"e_1_3_7_12_2","year":"2025","issue":null,"author":"Doshi R","volume":null,"first-page":null,"unstructured":"Doshi R. Gemini 3 Pro: The Frontier of Vision AI. The Keyword; 2025. Published. https://blog.google/technology/developers/gemini-3-pro-vision/. Accessed 27 December 2025.","article-title":null,"journal-title":null,"doi-asserted-by":null},{"DOI":"10.1038/sdata.2018.251","key":"e_1_3_7_13_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":null,"key":"e_1_3_7_14_2","year":null,"issue":null,"author":"Datta S","volume":null,"first-page":null,"unstructured":"Datta S, Buchireddygari D, Kaza LVC, Karnwal U, Bhatti HBS, Singh K. Gemini 3.0 Pro Surpasses Radiology Trainees on Radiology's Last Exam (Radle). CRASH Lab. Published November 20, 2025. https://drdattaaiims.github.io/Gemini-3.0-Radiology-2025.html. Accessed 27 December 2025.","article-title":null,"journal-title":null,"doi-asserted-by":null},{"DOI":"10.1148/radiol.232178","key":"e_1_3_7_15_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1038/s41746-025-02229-2","key":"e_1_3_7_16_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":null,"key":"e_1_3_7_17_2","year":null,"issue":null,"author":"Radiological Society of North America (RSNA)","volume":null,"first-page":null,"unstructured":"Radiological Society of North America (RSNA). RSNA 2024 Lumbar Spine Degenerative Classification Dataset. Kaggle. https://www.kaggle.com/competitions/rsna-2024-lumbar-spine-degenerative-classification. Accessed 17 March 2026.","article-title":null,"journal-title":null,"doi-asserted-by":null},{"DOI":"10.48550/arXiv.2506.09162","key":"e_1_3_7_18_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":"Richards TJ Flanders AE Colak E et al. 2025;The RSNA lumbar degenerative imaging Spine classification (LumbarDISC) dataset. arXiv preprint arXiv:2506.09162. doi:10.48550/arXiv.2506.09162.","article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1007/s00330-023-09483-6","key":"e_1_3_7_19_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3171/2025.4.Focus24670","key":"e_1_3_7_20_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1148/radiol.2021204289","key":"e_1_3_7_21_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1007/s00586-017-4956-3","key":"e_1_3_7_22_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3389/fendo.2022.890371","key":"e_1_3_7_23_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1177/21925682241299332","key":"e_1_3_7_24_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1007/s00586-023-08089-2","key":"e_1_3_7_25_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3390/diagnostics13182975","key":"e_1_3_7_26_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1007/s00330-024-11080-0","key":"e_1_3_7_27_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1101/2024.12.06.24318595","key":"e_1_3_7_28_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1007/s00586-025-09443-2","key":"e_1_3_7_29_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1038/nature14539","key":"e_1_3_7_30_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1016/j.media.2017.07.005","key":"e_1_3_7_31_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1148/radiol.2020192224","key":"e_1_3_7_32_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1056/NEJMp1714229","key":"e_1_3_7_33_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1148/radiol.2019191586","key":"e_1_3_7_34_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1038/s41591-020-1034-x","key":"e_1_3_7_35_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1109/tmi.2016.2535302","key":"e_1_3_7_36_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1371/journal.pmed.1002683","key":"e_1_3_7_37_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":null,"key":"e_1_3_7_38_2","year":"1990","issue":null,"author":"White AA","volume":null,"first-page":null,"unstructured":"White AA, Panjabi MM. Clinical Biomechanics of the Spine. 2nd ed. Philadelphia, PA: J. B. Lippincott; 1990. https://www.leomed.at/listhoscan/white_90.pdf. Accessed 27 December 2025.","article-title":null,"journal-title":null,"doi-asserted-by":null},{"DOI":"10.1109/tmi.2017.2673121","key":"e_1_3_7_39_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.21037/qims-23-1740","key":"e_1_3_7_40_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.3390/diagnostics13111947","key":"e_1_3_7_41_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":"Shi P Qiu J Abaxi SMD Wei H Lo FP Yuan W. Generalist vision foundation models for medical imaging: a case study of Segment Anything Model on zero-shot medical segmentation. Diagnostics (Basel). 2023;13(11):1947. doi:10.3390/diagnostics13111947. (In eng).","article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1016/j.landig.2025.100960","key":"e_1_3_7_42_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1016/j.jacr.2025.06.013","key":"e_1_3_7_43_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.2196/64290","key":"e_1_3_7_44_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.2196/68560","key":"e_1_3_7_45_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"},{"DOI":"10.1016/j.acra.2025.10.051","key":"e_1_3_7_46_2","year":null,"issue":null,"author":null,"volume":null,"first-page":null,"unstructured":null,"article-title":null,"journal-title":null,"doi-asserted-by":"publisher"}],"short-title":[],"update-policy":"https://doi.org/10.1177/sage-journals-update-policy","alternative-id":["10.1177/21925682261448823"],"article-number":"21925682261448823","content-domain":{"domain":["journals.sagepub.com"],"crossmark-restriction":true},"original-title":[],"container-title":["Global Spine Journal"],"reference-count":45,"published-online":{"date-time":null,"timestamp":null,"date-parts":[[2026,5,6]]},"references-count":45,"short-container-title":["Global Spine Journal"],"is-referenced-by-count":0},"unpaywall":{"doi":"10.1177/21925682261448823","is_oa":true,"title":"Benchmarking Multimodal Vision Frontier Models With Lumbar Spine MRIs for Grading Lumbar Spinal Stenosis","updated":"2026-05-12T08:54:10+00:00","published_date":"2026-05-06","best_oa_location":{"url":"https://doi.org/10.1177/21925682261448823","pmh_id":null,"url_for_pdf":null,"url_for_landing_page":"https://doi.org/10.1177/21925682261448823"}}},"created_at":"2026-05-14T13:22:58.227Z","updated_at":"2026-05-14T15:13:26.918Z","tags":[{"paper_doi":"10.1177/21925682261448823","tag_type":"age_group","value":"mixed","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"brand","value":"Gemini 3 Pro","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"category","value":"Clinical Neurology","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"condition","value":"Lumbar Spinal Stenosis","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"impact_flag","value":"safety_signal","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"level_of_evidence","value":"3","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"modality","value":"Multimodal AI / Large Language Model","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"sample_size","value":"n=100","value_numeric":100},{"paper_doi":"10.1177/21925682261448823","tag_type":"sample_size","value":"n=500","value_numeric":500},{"paper_doi":"10.1177/21925682261448823","tag_type":"study_design","value":"Diagnostic Accuracy Study","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"Benchmark interobserver agreement among readers was not reported.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"Gemini 3 Pro referenced in this paper (released late 2025) may differ from earlier public versions; version-specific performance cannot be independently verified from the paper alone.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"Model inputs were video montages of MRI series rather than typical PACS viewing, which may affect applicability.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"No p-values or confidence intervals were reported for between-arm comparisons despite noting non-significant differences.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"Sample size is 100 exams (500 disc levels); while powered for the primary endpoint, external validity to diverse populations and imaging protocols is limited.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"Severe class under-representation may inflate overall accuracy and obscure harmful misclassification.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"Sex of patients is not reported in the paper; the sex tag was omitted accordingly.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"The RSNA/ASNR consensus reference standard lacks reported interobserver agreement, so its reliability as a gold standard is uncertain.","value_numeric":null},{"paper_doi":"10.1177/21925682261448823","tag_type":"warning","value":"The study tested only a single commercial AI model (Gemini 3 Pro); results may not generalize to other frontier multimodal models or radiology-tuned systems.","value_numeric":null}],"summaries":{"what_studied":"A diagnostic accuracy study evaluated Google Gemini 3 Pro for grading lumbar spinal canal stenosis on video MRI using the RSNA 2024 Lumbar Spine Degenerative Classification Dataset with ASNR consensus labels, comparing performance with vs without localizer overlays across 100 exams (500 disc levels; 371 normal/mild, 74 moderate, 55 severe).","key_result":"Without localizer, accuracy was 75.6% (378/500) with fair agreement (κw = 0.39); severe stenosis sensitivity was 41.8%, with 43.6% of severe graded as normal/mild and 58.2% as non-severe. With localizer overlays, accuracy was 73.2% (366/500) with κw = 0.32; severe sensitivity fell to 30.9%, and 52.7% of severe were graded as normal/mild; differences were not significant.","limitations":"Interobserver agreement among contributing readers to the benchmark was not reported. Marked class imbalance (371/500 normal/mild) likely inflated accuracy, and the model was tested on video montages rather than native MRI viewers.","practice_takeaway":"Gemini 3 Pro missed many severe lumbar stenosis cases despite fair κw and mid‑70% accuracy; do not rely on it to rule in or rule out severe disease on volumetric MRI. Adding localizer overlays did not improve performance."}}],"initialOffset":1,"fetchConfig":{"kind":"tags","tags":[{"tag_type":"modality","value":"Multimodal AI / Large Language Model"}]},"breadcrumbItems":[{"name":"Home","href":"/"},{"name":"Research Hub","href":"/hub"},{"name":"Modalities","href":"/hub/tag/modality"},{"name":"Multimodal AI / Large Language Model"}],"hrefConfig":{"basePath":"/hub/papers","webViewQueryParams":{"hub_ctx":"tag","hub_tag_type":"modality","hub_tag_value":"Multimodal AI / Large Language Model"}}}]]

Papers