A new proof-of-concept study published online on 21 June 2026 in Multiple Sclerosis and Related Disorders examines microstructural integrity in juxtacortical U-fibers and its association with working memory performance in patients with multiple sclerosis. The research, led by Cristian Montalba along with Pamela Franco, Raúl Caulier-Cisterna, Juan Pablo Cruz, Claudia Cárcamo, Marcelo E Andia, and Ethel Ciampi, applies diffusion magnetic resonance imaging combined with machine learning techniques to detect subtle changes not visible on conventional scans.
Background on Multiple Sclerosis and Cognitive Challenges
Multiple sclerosis is a chronic autoimmune and neurodegenerative disorder marked by myelin destruction in the central nervous system, formation of lesions, and disruption of neuronal communication. Cognitive impairment appears as an early feature, affecting domains including executive function, learning and memory, complex attention, and information processing speed. Working memory deficits emerge early, impacting daily activities, quality of life, and employment. These changes often remain subtle and masked by cognitive reserve or fatigue, highlighting the need for objective measures.
Traditional focus has centered on long-range white matter tracts, yet the superficial or juxtacortical white matter, rich in short-range association U-fibers, has received less attention. These arch-shaped fibers connect neighboring cortical gyri and support local integration. Damage here may signal an early stage of juxtacortical disconnection preceding broader structural changes.
Understanding Key Assessment Tools and Imaging Metrics
The Paced Auditory Serial Addition Test, or PASAT, serves as a validated instrument for evaluating working memory, auditory information processing speed, flexibility, and calculation ability in multiple sclerosis. Participants add sequentially presented numbers, requiring rapid mental manipulation and retention of information. Performance on this test correlates with disease severity and structural damage, particularly in juxtacortical areas.
Fractional anisotropy, derived from diffusion tensor imaging, quantifies white matter integrity by measuring directional coherence of water diffusion along axonal fibers. Reductions in fractional anisotropy indicate demyelination and axonal loss. In multiple sclerosis, such reductions appear in tracts linked to cognitive performance, though the precise regional contributions require advanced analysis.
Photo by Bhautik Patel on Unsplash
Study Design and Machine Learning Framework
The investigation included 35 cognitively preserved healthy controls and 58 multiple sclerosis patients, encompassing both cognitively preserved and impaired individuals. Researchers extracted fractional anisotropy maps from 100 U-fiber regions. Working memory was quantified via PASAT Z-scores adjusted for age and education.
Fifteen regression models were trained alongside sequential forward selection to estimate PASAT Z-scores from the fractional anisotropy features. Performance assessment relied on 5-fold nested cross-validation, using mean squared error, mean absolute error, and coefficient of determination as metrics. This approach handles high-dimensional neuroimaging data and captures non-linear relationships between regional integrity and cognitive outcomes.
Key Findings from the Analysis
Tree-based ensemble models demonstrated superior performance. The CatBoost model achieved a mean squared error of 0.256 plus or minus 0.068, mean absolute error of 0.421 plus or minus 0.068, and coefficient of determination of 0.488 plus or minus 0.128. Sequential forward selection identified sixteen fractional anisotropy features as most informative, with the left superior temporal and right inferior parietal regions ranking highest in importance.
These results indicate that diffusion magnetic resonance imaging-derived fractional anisotropy measures from short-range U-fibers can estimate working memory performance with clinically meaningful accuracy. The findings support the potential of these features to detect early juxtacortical dysconnectivity as a biomarker for cognitive monitoring.
Implications for Early Detection and Monitoring
The study underscores how machine learning can reveal patterns in normal-appearing superficial white matter that traditional methods overlook. By ranking feature importance, the analysis maps topographical distributions of damage linked to cognitive decline. This non-invasive pipeline offers a pathway toward sensitive imaging markers for risk stratification in the early stages of multiple sclerosis.
Stakeholders including neurologists, radiologists, and researchers may find value in integrating such approaches into longitudinal studies. Patients could benefit from earlier identification of connectivity changes that precede overt disability.
Future Directions and Broader Context
Further validation in larger cohorts and across disease subtypes will strengthen the generalizability of these observations. Integration with other modalities, such as functional imaging or additional cognitive batteries, could refine predictive models. The proof-of-concept nature emphasizes exploratory insights rather than immediate clinical deployment.
Academic institutions and research centers continue to advance neuroimaging and computational methods that address the multifaceted nature of multiple sclerosis. This work contributes to ongoing efforts to link brain structure with functional outcomes through data-driven techniques.
Readers interested in related career opportunities in neuroimaging research or multiple sclerosis studies may explore positions listed on academic job platforms.
Accessing the Original Publication
The full article appears in Multiple Sclerosis and Related Disorders and is available at https://www.sciencedirect.com/science/article/abs/pii/S221103482600369X. An additional repository entry is hosted at https://repositorio.uc.cl/handle/11534/109374. The journal site provides further context on recent publications in the field at https://www.msard-journal.com/inpress.
