Helping Researchers See Alzheimer’s Before It Starts - Profile Zheyu Wen

What if we could detect neurodegenerative diseases before symptoms ever appear? Conditions such as Alzheimer’s can quietly take root long before a diagnosis is made, gradually transforming daily life into unfamiliar and often frightening territory for patients and their loved ones. But researchers are working to change that, and graduate student Zheyu Wen is at the forefront, using technology and computational mechanics to drive new developments.

Wen is a Ph.D. student in the Computational Science, Engineering, and Mathematics (CSEM) Program at the Oden Institute for Computational Engineering and Sciences where he is a member of the Parallel Algorithms for Data Analysis and Simulation Group (PADAS).  His research combines engineering and medicine to develop computational models that analyze how the brain changes over time.

Originally from China, Wen earned his bachelor’s degree in communication engineering from the University of Electronic Science and Technology of China, and completed his master’s degree in electrical and computer engineering from the University of Michigan, Ann Arbor.

"I don't have a personal connection to Alzheimer’s,“ he explained, “but I started doing brain-related research during my master’s degree, focusing on magnetic resonance imaging (MRI) reconstruction from subsampled data which was my first exposure to this field.”

After developing proficiency in neuroimaging, Wen came to The University of Texas at Austin to pursue his Ph.D. and joined the PADAS Group, led by professor George Biros, professor of mechanical engineering and Oden Institute Principal Faculty member. “Professor Biros had done work in brain tumor modeling and had recently started Alzheimer’s-related research. My background aligned well, and I also wanted to explore inverse problems and apply high-performance computing in this space.”

The goal is to diagnose Alzheimer’s sooner by identifying patterns in brain imaging linked to protein misfolding, a biological process in which proteins lose their proper structure and can no longer perform their normal function, sometimes clumping together into toxic buildups in the brain often before clinical symptoms appear. While most medical models focus on individual-level predictions, Wen’s approach is cohort-based. He builds models from a broader population, aiming to learn common spatiotemporal patterns in the progression of brain deterioration and uncover the mechanisms behind misfolded proteins and their harmful accumulations.

“Instead of making the model personal, we try to find common patterns that happen across the cohort. This is where AI and machine learning become very powerful,” Wen said. 

Traditional machine learning models often rely on raw patient data, which can include noisy data and outliers that could potentially skew predictions. Wen’s approach filters out those inconsistencies by identifying common patterns across large populations, resulting in models that are more meaningful and reliable.

Protein misfolding doesn’t just appear in one place. Instead, it progresses through the brain over time. Wen’s model captures this dynamic, giving researchers a fuller picture of when and where changes are happening, which may one day help inform treatment decisions. Interpretability is one of the strengths of Wen’s approach, as his model not only makes predictions, but also offers insight into why and how the brain is changing, a critical component for building trust in AI.

Wen's research is getting noticed with recent accolades for his work. He was one of two recipients of the 2024 Best Paper Award from the Medical Image Computing and Computer Assisted Intervention (MICCAI), given at the premier international conference on AI in medical imaging. Out of over 3,000 submissions, approximately 720 were accepted, and only two selected in the Best Paper category. The paper, which introduced a cohort-level model for tracking protein progression across brain regions, was praised for both its predictive accuracy and biological interpretability.

In addition, Wen was one of two researchers who received the inaugural Paul A. Navrátil Young Researcher Award, which recognizes outstanding UT students for their innovative use of Texas Advanced Computing Center (TACC) resources, cross-disciplinary excellence, and advancing knowledge.

Processing large imaging datasets and training machine learning models at scale is no small feat, and this is where TACC comes in. The high-performance computing resources offered at TACC allow Wen to analyze vast amounts of brain imaging data with both speed and precision.

By using large datasets of brain scans from Alzheimer’s patients, collected through MRI and positron emission tomography (PET), Wen zeroes in on the spatial and temporal accumulation of tau and amyloid proteins, two key biomarkers of the disease. This allows Wen’s model to provide a broader view than traditional patient-specific models.

“We were able to train our model on TACC clusters,” he explained, “which is important because our model is large-scale, and the neuroimaging data itself is huge.” While his current focus is Alzheimer's, Wen notes that this framework could eventually be adapted to study other neurodegenerative diseases, such as Parkinson’s, where protein misfolding and progression patterns play a similar key role.

With a background spanning electrical engineering and computational neuroscience, Wen is carving out a distinct niche at the intersection of data science and neuroimaging. His recent awards demonstrate not only technical innovation, but also the broader promise of machine learning in advancing our understanding of neurodegenerative disease.