Project: Data-Driven Dimensionality Reduction and Interpretation of Diffusion MRI Signals in Healthy and Tumor-Affected Brain Tissue
Description
This project aims to develop a data-driven framework for analyzing diffusion MRI (dMRI) signals in healthy and tumor-affected brain tissue using dimensionality reduction techniques such as t-SNE, UMAP, and autoencoders. By embedding the complex, multi-dimensional dMRI signal into a low-dimensional space (e.g., 2–3 components per voxel), the goal is to generate interpretable maps that capture intrinsic microstructural features of brain tissue without relying on traditional biophysical models. These component maps will be quantitatively compared with established dMRI metrics to assess how well data-driven representations reflect known microstructural properties.
The study will further test whether low-dimensional representations derived from simpler acquisitions (e.g., single-shell) can approximate metrics that typically require advanced multi-shell data. Finally, the approach will be applied to individual tumor cases to explore how different components reflect tissue heterogeneity—distinguishing tumor core, edema, and infiltrated regions. The outcome will be a validated framework offering new ways to visualize, interpret, and simplify diffusion MRI data in both research and clinical contexts.
Details
- Student
-
JWJasper Wilfling
- Supervisor
-
Maxime Chamberland
- Secondary supervisor
-
Rembrandt Bakker