B. Ng, G. Varoquaux, J.B. Poline, B. Thirion

Despite that neural dynamics are largely constrained by the underlying fiber pathways, such anatomical information is rarely exploited in estimating functional connectivity. To integrate DWI and fMRI, we propose a novel approach based on sparse Gaussian graphical models. The implications of incorporating connectivity learned from RS-fMRI and DWI data into task activation detection and sparse classifier learning are also explored.

B. Ng, G. Hamarneh, R. Abugharbieh

The strong noise in fMRI data greatly hampers reliable activation detection. We propose exploiting the integrative property of the brain as well as commonalities across subjects within a group to improve activation detection. State-of-the-art graph-based techniques are extended and explored for incorporation of these priors.

B. Ng, M.J. McKeown, R. Abugharbieh

Clinical studies often focus on analyzing specific regions of interest (ROIs) by averaging the signals within the ROIs. The spatial pattern of regional activity is thus ignored. To mitigate this limitation, we propose novel invariant spatial descriptors for capturing the spatial information encoded in regional activity patterns.

B. Ng, M.J. McKeown, R. Abugharbieh

The high inter-subject variability renders establishment of network correspondence across subjects difficult. We propose incorporating group information into sparse network detection to alleviate this complication. The gain of combining sparsity and group priors in brain network identification are examined.

B. Ng, G. Hamarneh, R. Abugharbieh

The high dimensionality of fMRI data given the typically limited number of noisy samples poses a major challenge to classifier learning. To constrain this ill-conditioned problem, we propose a novel classifier learning formulation that facilitates joint integration of spatiotemporal priors and structured sparsity. The benefits of modeling the intrinsic spatiotemporal properties of brain activity in addition to enforcing sparsity are investigated.

Multimodal Integration for Neuroimage Analysis

Sparse Clustering for Network Identification

Spatial Characterization of Brain Activity Patterns

Prior-informed Brain Activation Detection

Sparse Classifier Learning for Brain Decoding

My principal research interests lie in developing new computational methods based on machine learning, sparse optimization, graphical models, and Bayesian statistics for biomedical applications. I am currently investigating new techniques for multimodal neuroimaging integration to improve brain connectivity inference. My future research direction is to additionally incorporate genetic and behavioural factors for establishing richer disease characterizations in enhancing disease detection and diagnosis.