{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,4,21]],"date-time":"2026-04-21T04:45:38Z","timestamp":1776746738722,"version":"3.51.2"},"reference-count":46,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,4,21]],"date-time":"2026-04-21T00:00:00Z","timestamp":1776729600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Neurol."],"abstract":"<jats:sec>\n                    <jats:title>Background and objectives<\/jats:title>\n                    <jats:p>Identifying multiple sclerosis (MS) in children early is critical, as early therapeutic intervention can improve outcomes. The anterior visual pathway has been demonstrated to be of central importance in diagnostic considerations for MS and has recently been identified as a fifth topography in the McDonald Diagnostic Criteria for MS. Optical coherence tomography (OCT) provides high-resolution retinal imaging and reflects the structural integrity of the retinal nerve fiber and ganglion cell inner plexiform layers. Whether multimodal deep learning models can use OCT alone to diagnose pediatric onset MS (POMS) is unknown.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We analyzed 3D OCT scans collected prospectively through the Neuroinflammatory Registry of the Hospital for Sick Children (REB#1000005356). Raw macular and optic nerve head images, and 52 automatically segmented features were included. We evaluated three classification approaches: (1) deep learning models (e.g., ResNet, DenseNet) for representation learning followed by classical ML classifiers, (2) ML models trained on OCT-derived features, and (3) multimodal models combining both via early and late fusion.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      Scans from individuals with POMS (onset 16.0\u202f\u00b1\u202f3.1\u202fyears, 51.0% female; 211 scans) and 29 children with non-inflammatory neurological conditions (13.1\u202f\u00b1\u202f4.0\u202fyears, 69.0% female, 52 scans) were included. The early fusion model achieved the highest performance (AUC: 0.90, weighted\n                      <jats:italic>F<\/jats:italic>\n                      <jats:sub>1<\/jats:sub>\n                      : 0.87, macro\n                      <jats:italic>F<\/jats:italic>\n                      <jats:sub>1<\/jats:sub>\n                      : 0.77, accuracy: 87%), outperforming both unimodal and late fusion models. The best unimodal feature-based model (SVC) yielded an AUC of 0.84, weighted\n                      <jats:italic>F<\/jats:italic>\n                      <jats:sub>1<\/jats:sub>\n                      of 0.85, macro\n                      <jats:italic>F<\/jats:italic>\n                      <jats:sub>1<\/jats:sub>\n                      of 0.73, and accuracy of 85%, while the best image-based model (ResNet101 with SVC) achieved an AUC of 0.79, weighted\n                      <jats:italic>F<\/jats:italic>\n                      <jats:sub>1<\/jats:sub>\n                      of 0.84, macro\n                      <jats:italic>F<\/jats:italic>\n                      <jats:sub>1<\/jats:sub>\n                      of 0.70, and accuracy of 87%. Late fusion underperformed, reaching 82% accuracy but failing in the minority class.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Discussion<\/jats:title>\n                    <jats:p>Multimodal learning with early fusion significantly enhances diagnostic performance by combining spatial retinal information with clinically relevant structural features. This approach captures complementary patterns associated with MS pathology and shows promise as an AI-driven tool to support pediatric neuroinflammatory diagnosis.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fneur.2026.1791696","type":"journal-article","created":{"date-parts":[[2026,4,21]],"date-time":"2026-04-21T04:22:00Z","timestamp":1776745320000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Multimodal machine learning for distinguishing pediatric multiple sclerosis from non-inflammatory conditions using optical coherence tomography"],"prefix":"10.3389","volume":"17","author":[{"given":"Chaojun","family":"Chen","sequence":"first","affiliation":[{"name":"Program in Neuroscience and Mental Health, SickKids Research Institute, The Hospital for Sick Children","place":["Toronto, ON, Canada"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Sahar","family":"Soltanieh","sequence":"additional","affiliation":[{"name":"Program in Neuroscience and Mental Health, SickKids Research Institute, The Hospital for Sick Children","place":["Toronto, ON, Canada"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Sajith","family":"Rajapaksa","sequence":"additional","affiliation":[{"name":"Program in Neuroscience and Mental Health, SickKids Research Institute, The Hospital for Sick Children","place":["Toronto, ON, Canada"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Farzad","family":"Khalvati","sequence":"additional","affiliation":[{"name":"Program in Neuroscience and Mental Health, SickKids Research Institute, The Hospital for Sick Children","place":["Toronto, ON, Canada"]},{"name":"Department of Medical Imaging, University of Toronto","place":["Toronto, ON, Canada"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"E. Ann","family":"Yeh","sequence":"additional","affiliation":[{"name":"Program in Neuroscience and Mental Health, SickKids Research Institute, The Hospital for Sick Children","place":["Toronto, ON, Canada"]},{"name":"Division of Neurology, Department of Pediatrics, The Hospital for Sick Children, University of Toronto","place":["Toronto, ON, Canada"]}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"1965","published-online":{"date-parts":[[2026,4,21]]},"reference":[{"key":"ref1","article-title":"Multiple sclerosis","year":"2025"},{"key":"ref2","doi-asserted-by":"publisher","first-page":"45","DOI":"10.1007\/s40263-021-00887-w","article-title":"Multiple sclerosis in children: differential diagnosis, prognosis, and disease-modifying treatment","volume":"36","author":"Jakimovski","year":"2022","journal-title":"CNS Drugs"},{"key":"ref3","doi-asserted-by":"publisher","first-page":"844","DOI":"10.1136\/jnnp-2017-317509","article-title":"Time matters in multiple sclerosis: can early treatment and long-term follow-up ensure everyone benefits from the latest advances in multiple sclerosis","volume":"89","author":"Cerqueira","year":"2018","journal-title":"J Neurol Neurosurg Psychiatry"},{"key":"ref4","doi-asserted-by":"publisher","first-page":"S53","DOI":"10.1212\/WNL.0000000000002822","article-title":"Pediatric optic neuritis","volume":"87","author":"Yeh","year":"2016","journal-title":"Neurology"},{"key":"ref5","doi-asserted-by":"publisher","first-page":"103761","DOI":"10.1016\/j.msard.2022.103761","article-title":"Progressive retinal changes in pediatric multiple sclerosis","volume":"61","author":"Longoni","year":"2022","journal-title":"Mult Scler Relat Disord"},{"key":"ref6","doi-asserted-by":"publisher","first-page":"441","DOI":"10.1016\/j.jaapos.2011.05.020","article-title":"Pediatric optic neuritis and risk of multiple sclerosis: meta-analysis of observational studies","volume":"15","author":"Waldman","year":"2011","journal-title":"J AAPOS"},{"key":"ref7","article-title":"Guidelines for MS diagnosis: McDonald criteria","author":"Wexler","year":"2025"},{"key":"ref8","doi-asserted-by":"publisher","first-page":"e00631","DOI":"10.1016\/j.neurot.2025.e00631","article-title":"Pediatric multiple sclerosis: improving outcome through high-efficacy therapies","volume":"22","author":"Aljomah","year":"2025","journal-title":"Neurotherapeutics"},{"key":"ref9","doi-asserted-by":"publisher","first-page":"2253","DOI":"10.1177\/13524585221112605","article-title":"Machine learning classification of multiple sclerosis in children using optical coherence tomography","volume":"28","author":"Ciftci Kavaklioglu","year":"2022","journal-title":"Mult Scler J"},{"key":"ref10","doi-asserted-by":"publisher","first-page":"451","DOI":"10.1111\/j.1755-3768.2011.02134.x","article-title":"RNFL thickness in MS-associated acute optic neuritis using SD-OCT: critical interpretation and limitations","volume":"89","author":"Serbecic","year":"2011","journal-title":"Acta Ophthalmol"},{"key":"ref11","doi-asserted-by":"publisher","first-page":"675","DOI":"10.3389\/fneur.2017.00675","article-title":"The temporal retinal nerve fiber layer thickness is the most important optical coherence tomography estimate in multiple sclerosis","volume":"8","author":"Birkeldh","year":"2017","journal-title":"Front Neurol"},{"key":"ref12","doi-asserted-by":"publisher","first-page":"574","DOI":"10.1016\/S1474-4422(16)00068-5","article-title":"Retinal thickness measured with optical coherence tomography and risk of disability worsening in multiple sclerosis: a cohort study","volume":"15","author":"Martinez-Lapiscina","year":"2016","journal-title":"Lancet Neurol"},{"key":"ref13","doi-asserted-by":"publisher","first-page":"2147","DOI":"10.1212\/WNL.0000000000001046","article-title":"Functional\u2013structural correlations in the afferent visual pathway in pediatric demyelination","volume":"83","author":"Yeh","year":"2014","journal-title":"Neurology"},{"key":"ref14","doi-asserted-by":"publisher","first-page":"802","DOI":"10.1177\/1352458509104586","article-title":"Retinal nerve fiber thickness in inflammatory demyelinating diseases of childhood onset","volume":"15","author":"Yeh","year":"2009","journal-title":"Mult Scler J"},{"key":"ref15","doi-asserted-by":"publisher","first-page":"537","DOI":"10.1146\/annurev.bioeng.8.061505.095802","article-title":"Machine learning for detection and diagnosis of disease","volume":"8","author":"Sajda","year":"2006","journal-title":"Annu Rev Biomed Eng"},{"key":"ref16","first-page":"281","article-title":"A review on machine learning classification techniques for plant disease detection","author":"Shruthi","year":"2019"},{"key":"ref17","doi-asserted-by":"publisher","first-page":"541","DOI":"10.3390\/healthcare10030541","article-title":"Machine-learning-based disease diagnosis: a comprehensive review","volume":"10","author":"Ahsan","year":"2022","journal-title":"Healthcare"},{"key":"ref18","doi-asserted-by":"publisher","first-page":"48","DOI":"10.1038\/s41746-022-00592-y","article-title":"Machine learning for medical imaging: methodological failures and recommendations for the future","volume":"5","author":"Varoquaux","year":"2022","journal-title":"npj Digit Med"},{"key":"ref19","doi-asserted-by":"publisher","first-page":"60","DOI":"10.1109\/MPUL.2011.942929","article-title":"Medical image analysis","volume":"2","author":"Ritter","year":"2011","journal-title":"IEEE Pulse"},{"key":"ref20","doi-asserted-by":"publisher","first-page":"116","DOI":"10.1109\/RBME.2020.3007816","article-title":"Machine learning for clinical outcome prediction","volume":"14","author":"Shamout","year":"2020","journal-title":"IEEE Rev Biomed Eng"},{"key":"ref21","doi-asserted-by":"publisher","first-page":"31","DOI":"10.1177\/00031348221109478","article-title":"Predicting patient-reported outcomes following surgery using machine learning","volume":"89","author":"Hassan","year":"2023","journal-title":"Am Surg"},{"key":"ref22","doi-asserted-by":"publisher","first-page":"1675","DOI":"10.30574\/wjarr.2024.21.2.0641","article-title":"Machine learning\u2019s role in personalized medicine & treatment optimization","volume":"21","author":"Kumar","year":"2024","journal-title":"World J Adv Res Rev"},{"key":"ref23","doi-asserted-by":"publisher","first-page":"100704","DOI":"10.1016\/j.oor.2024.100704","article-title":"AI-driven diagnostics and personalized treatment planning in oral oncology: innovations and future directions","volume":"13","author":"Satheeskumar","year":"2025","journal-title":"Oral Oncol Rep"},{"key":"ref24","doi-asserted-by":"publisher","first-page":"505","DOI":"10.1148\/rg.2017160130","article-title":"Machine learning for medical imaging","volume":"37","author":"Erickson","year":"2017","journal-title":"Radiographics"},{"key":"ref25","doi-asserted-by":"publisher","first-page":"242","DOI":"10.1016\/j.ejmp.2021.04.016","article-title":"Artificial intelligence and machine learning for medical imaging: a technology review","volume":"83","author":"Barrag\u00e1n-Montero","year":"2021","journal-title":"Phys Med"},{"key":"ref26","doi-asserted-by":"publisher","first-page":"998","DOI":"10.1002\/jmri.26852","article-title":"Machine learning in breast MRI","volume":"52","author":"Reig","year":"2020","journal-title":"J Magn Reson Imaging"},{"key":"ref27","doi-asserted-by":"publisher","first-page":"331","DOI":"10.1007\/s40471-018-0165-9","article-title":"Using machine learning to identify health outcomes from electronic health record data","volume":"5","author":"Wong","year":"2018","journal-title":"Curr Epidemiol Rep"},{"key":"ref28","doi-asserted-by":"publisher","first-page":"1419","DOI":"10.1093\/jamia\/ocy068","article-title":"Opportunities and challenges in developing deep learning models using electronic health records data: a systematic review","volume":"25","author":"Xiao","year":"2018","journal-title":"J Am Med Inform Assoc"},{"key":"ref29","doi-asserted-by":"publisher","first-page":"1544","DOI":"10.1001\/jamainternmed.2018.3763","article-title":"Potential biases in machine learning algorithms using electronic health record data","volume":"178","author":"Gianfrancesco","year":"2018","journal-title":"JAMA Intern Med"},{"key":"ref30","doi-asserted-by":"publisher","first-page":"233","DOI":"10.1016\/j.future.2018.12.041","article-title":"Automated detection of cancerous genomic sequences using genomic signal processing and machine learning","volume":"98","author":"Liu","year":"2019","journal-title":"Futur Gener Comput Syst"},{"key":"ref31","doi-asserted-by":"publisher","first-page":"325","DOI":"10.1093\/bioinformatics\/btab681","article-title":"Prediction of antimicrobial resistance based on whole-genome sequencing and machine learning","volume":"38","author":"Ren","year":"2022","journal-title":"Bioinformatics"},{"key":"ref32","first-page":"26140","article-title":"Calibrating multi-modal representations: a pursuit of group robustness without annotations","author":"You","year":"2024"},{"key":"ref33","doi-asserted-by":"publisher","first-page":"789","DOI":"10.1109\/JBHI.2023.3331626","article-title":"In-distribution and out-of-distribution self-supervised ECG representation learning for arrhythmia detection","volume":"28","author":"Soltanieh","year":"2023","journal-title":"IEEE J Biomed Health Inform"},{"key":"ref34","doi-asserted-by":"publisher","first-page":"710","DOI":"10.1109\/TBME.2018.2854899","article-title":"Robust heartbeat detection from multimodal data via CNN-based generalizable information fusion","volume":"66","author":"Chandra","year":"2018","journal-title":"IEEE Trans Biomed Eng"},{"key":"ref35","doi-asserted-by":"publisher","first-page":"34823","DOI":"10.1371\/journal.pone.0034823","article-title":"The OSCAR-IB consensus criteria for retinal OCT quality assessment","volume":"7","author":"Tewarie","year":"2012","journal-title":"PLoS One"},{"key":"ref36","doi-asserted-by":"publisher","first-page":"8323","DOI":"10.1167\/iovs.11-7962","article-title":"Macular ganglion cell\u2013inner plexiform layer: automated detection and thickness reproducibility with spectral domain\u2013optical coherence tomography in glaucoma","volume":"52","author":"Mwanza","year":"2011","journal-title":"Invest Ophthalmol Vis Sci"},{"key":"ref37","doi-asserted-by":"publisher","first-page":"5","DOI":"10.1023\/a:1010933404324","article-title":"Random forests","volume":"45","author":"Breiman","year":"2001","journal-title":"Mach Learn"},{"key":"ref38","doi-asserted-by":"publisher","first-page":"367","DOI":"10.1016\/s0167-9473(01)00065-2","article-title":"Stochastic gradient boosting","volume":"38","author":"Friedman","year":"2002","journal-title":"Comput Stat Data Anal"},{"key":"ref39","first-page":"785","article-title":"XGBoost: a scalable tree boosting system","author":"Chen","year":"2016"},{"key":"ref40","doi-asserted-by":"publisher","first-page":"1883","DOI":"10.4249\/scholarpedia.1883","article-title":"K-nearest neighbor","volume":"4","author":"Peterson","year":"2009","journal-title":"Scholarpedia"},{"key":"ref41","doi-asserted-by":"publisher","first-page":"18","DOI":"10.1109\/5254.708428","article-title":"Support vector machines","volume":"13","author":"Hearst","year":"1998","journal-title":"IEEE Intell Syst Appl"},{"key":"ref42","doi-asserted-by":"crossref","first-page":"81","DOI":"10.1007\/978-1-4615-0377-4_5","article-title":"Artificial neural network","volume-title":"Interdisciplinary Computing in Java Programming","author":"Wang","year":"2003"},{"key":"ref43","first-page":"3347","article-title":"Transfusion: understanding transfer learning for medical imaging","author":"Raghu","year":"2019"},{"key":"ref44","first-page":"1","article-title":"Analysis of augmentations for contrastive ECG representation learning","author":"Soltanieh","year":"2022"},{"key":"ref45","doi-asserted-by":"publisher","first-page":"1299","DOI":"10.1109\/TMI.2016.2535302","article-title":"Convolutional neural networks for medical image analysis: full training or fine tuning","volume":"35","author":"Tajbakhsh","year":"2016","journal-title":"IEEE Trans Med Imaging"},{"key":"ref46","doi-asserted-by":"publisher","first-page":"115","DOI":"10.1038\/nature21056","article-title":"Dermatologist-level classification of skin cancer with deep neural networks","volume":"542","author":"Esteva","year":"2017","journal-title":"Nature"}],"container-title":["Frontiers in Neurology"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fneur.2026.1791696\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,4,21]],"date-time":"2026-04-21T04:22:02Z","timestamp":1776745322000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fneur.2026.1791696\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,4,21]]},"references-count":46,"alternative-id":["10.3389\/fneur.2026.1791696"],"URL":"https:\/\/doi.org\/10.3389\/fneur.2026.1791696","relation":{},"ISSN":["1664-2295"],"issn-type":[{"value":"1664-2295","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,4,21]]},"article-number":"1791696"}}