Frontotemporal Dementia Research
Making machine-aided diagnosis explainable to doctors.
Frontotemporal dementia, a specific type of dementia, is extremely hard for doctors to diagnose. A highly accurate model means little if doctors cannot interpret or trust its decisions. Under the guidance of Dr. Shihab Shamma and Maryam Shaghaghi at UMD, I built a system that highlights abnormal brain activity behind each prediction. Instead of a black box, doctors can see why the model made its decision.
For this research, I was recognized by Regeneron Pharmaceuticals as a STS Top 300 Scholar.
Year
2024
Role
Research Intern
Team
Bryan: Research Intern, and the wonderful lab members
Tools
Python, scikit-learn, NumPy, Discrete Wavelet Transform
Raw brain activity recorded from patients is messy and noisy. Each line is electrical signals from a different part of the brain.
I worked on a pipeline that transformed raw brain signals into measurable features across regions and frequency bands, then visualized how frontotemporal dementia patients differ from healthy controls.
Visualization of gamma and beta activity across brain regions in FTD patients. Different statistical measures reveal region-specific signal patterns.
t-SNE projection of extracted features. Left: overlapping clusters with noisy features. Right: distinct groupings after isolating the most informative brain signals.