Yikang Li

My current project explores the underrecognized role of white matter (WM) blood-oxygen-level dependent (BOLD) signals in preclinical Alzheimer's disease (AD). While most fMRI studies focus on gray matter (GM), WM may provide complementary information. We developed BrainVAE, a transformer-based variational autoencoder that integrates WM and GM functional connectivity through a three-stage training pipeline. Compared with nine benchmark machine learning and deep learning models, BrainVAE achieved superior accuracy and robustness, especially when combining WM and GM inputs. Attention-based interpretability further identified biologically meaningful WM bundles, highlighting WM BOLD as a valuable biomarker for early AD detection.

My research focuses on leveraging machine learning algorithms to explore the role of white matter BOLD signals in brain function.