The model, called div-mDCSRN-Flow, was created to enhance the resolution of 4D flow MRI scans. These scans are used to track cerebrospinal fluid and blood movement throughout the brain, helping researchers better understand how these flows change in both healthy and diseased states. The work was led by Neal Patel, Emily Bartusiak, Sean Rothenberger, PhD, A.J. Schwichtenberg, Edward Delp and Vitaliy Rayz, Professor of Biomedical Engineering at the Weldon School.

Traditional 4D flow MRI is limited in resolution, often making it difficult to capture small-scale or subtle changes in neurofluid dynamics. To address this challenge, the team trained their model using synthetic MRI data from both healthy and Alzheimer’s cases. The result is a physics-informed neural network that combines machine learning with fluid dynamics principles to maintain accuracy while enhancing image detail.

In testing, the new method demonstrated a 22.5% reduction in error rates in core brain regions and a 49.5% reduction at the edges of images. It also proved effective in simulations, lab experiments and real-world imaging scenarios.

The research reflects the strength of interdisciplinary collaboration at Purdue, bringing together expertise from the Weldon School of Biomedical Engineering, the College of Engineering, the Elmore Family School of Electrical and Computer Engineering, the College of Health and Human Sciences and the School of Mechanical Engineering.

As researchers continue to refine diagnostic tools and explore how fluid dynamics relate to brain health, innovations like div-mDCSRN-Flow may help lead the way toward earlier detection and improved understanding of neurological disorders.