Skip to content. Skip to main navigation.

Faculty Highlight

Dajiang Zhu

Dajiang Zhu

Dajiang Zhu, Ph.D., is an Assistant Professor at the CSE Department. His research focuses on Brain Imaging Computing, Computational Neuroscience and Big Data solutions for medical data analysis through computational modeling and machine learning methods. A major goal of his research is to understand the organizational architecture of human brain functions and its relationship to brain structures.

Malware

In brain science, a challenging issue that has received intensive attention recently is how to establish a stable and neuroscience grounded foundation or substrates for synthetically and quantitatively measuring connectivity and dynamic interactions, either within individual brains or comparing them across populations. Due to the scarcity of ground-truth data, researchers have to validate and replicate data from multiple subjects so that sufficient statistical power can be achieved. However, this population-level data pooling step requires the determination of accurate correspondences between regions of interest (ROIs) across different brains, which is a major barrier in human brain mapping and neuroimaging for several decades. This challenge comes from not only the remarkable individual variability of cortical anatomy and connection, but also the critical lack of effective computational model for robustly and comprehensively estimating the brain structural connectivity patterns. To address this challenge, Dr. Zhu proposed a framework to explore the most consistent (common) structural connectivity patterns (Fig.1) across different populations and this framework is well known as Dense Individualized and Common Connectivity-based Cortical landmarks (DICCCOL) in the brain imaging and computational neuroscience field.



Binary Code

DICCCOL is a sparse atlas of human brain based on consistency of white matter pathways. Through DICCCOL, researchers can easily construct structural or functional networks from a few hundreds of key nodes in the brain to perform further comparison or group analysis (Fig.2). DICCCOL prediction toolkit has been downloaded more than 300 times after it was first published in 2013. In addition, DICCCOL framework has been applied on different brain diseases to examining their structural and functional abnormalities, such as Schizophrenia (SZ), Mild Cognitive Impairment (MCI), Attention-Deficit Hyperactivity Disorder (ADHD) and Post-Traumatic Stress Disorder (PTSD). In comparison with traditional methods for brain structural/functional connectivity analysis that rely on the Brodmann (Atlas) map or functional parcellation map, DICCCOL-based connectomes can offer finer granularity, better functional homogeneity, more accurate functional localization, and automatically-established cross-subjects correspondences.

Malware

Dr. Zhu’s current research focuses on developing novel machine learning methods for functional magnetic resonance imaging (fMRI) analysis. Traditionally, the subtraction approach (contrast between task and baseline epochs) has been the dominant methodology in both task-based fMRI paradigm design and fMRI data analysis, based on which a majority of previous human neuroimaging/brain mapping studies and conclusions were derived. Despite the remarkable successes and significant neuroscientific insights achieved by the subtraction approach, it has difficulty in reconstructing concurrent, interacting functional networks, as it was already recognized and pointed out in the literature that spatially overlapping networks subserving different functions are likely to be unnoticed by the blocked subtraction paradigms. In order to address this fundamental question and bridge the current significant knowledge gaps, Dr. Zhu is working on an innovative computational framework of sparse coding of whole-brain fMRI signals.One advantage of using sparse learning on brain fMRI signals is that it can effectively and robustly uncover multiple functional networks (Fig. 3), including both task-evoked and resting state networks (RSNs), simultaneously.