Wenjin Zhou

My research passion lies in interdisciplinary and collaborative research on data science, interactive visualization, data-intensive computing, and information retrieval for brain and health sciences. I believe that the impact of application-driven contributions to expand computer science to interdisciplinary research will become more and more important in exploring and understanding health and scientific problems.

The diagram above outlines the three research areas that I work on. The first is the computational solutions for extraction of ``hidden'' information from various biomedical and scientific disciplines through mathematical modeling and analysis. The second area explores powerful visualization tools for combining multi-dimensional information obtained from various data format and imaging modalities. The third area attacks the problem of designing efficient interactive user interfaces for understanding and exploring scientific data at multiple scales. These three research areas interact with one another to expand our knowledge in computer science, brain and health sciences, and lead to new discoveries and to advances in healthcare.

In particular, I have been focusing on developing computational techniques and interactive visualization applications through close collaboration with neuropsychologists, including Dr. Stephen Correia at Butler Hospital and Dr. David Tate at Harvard Medical School, to improve understanding of the brain and solve medical problems. I have developed new computational approaches towards the virtual histology of brain microstructure using diffusion imaging to provide reliable and sensitive biomarkers for brain disease diagnosis and prevention. The information extraction aspect of the work is meaningful in many other health-related areas of life sciences such as biology, DNA sequencing and cancer genomics, which I plan to explore. I have also designed interactive visualization interfaces for exploring multidimensional brain data in order to help brain scientists test their hypothesis on brain segmentation, connectivity and pathological changes. I aim to investigate the design of powerful visualizations that encode multidimensional information and efficient interactive interfaces that explores multiscale data across various scientific domains.

Here's a list of our publication.

1. Data Intensive Computing: Computational modeling and image analysis for information extraction

Brain scientists obtain numerous types of medical images in order to understand the exceedingly complex structure and functions of the brain. Although these images encode important brain structural information, they are only relative quantities measured in the physical environment of the imaging process. More direct measurements of the brain properties remain 'hidden'' in those data. Over the next three years, I aim to contribute to the extraction of these important physical quantities in brain science, working closely with brain scientists at nearby medical centers. The process usually requires physical analysis of the underlying environment, mathematical modeling, and robust computational algorithms. Over the next decade, I intend to pursue similar computational information extraction problems in other areas of health sciences such as biology, DNA sequencing and cancer genomics through close collaborations with researchers in these areas.

1.1 Brain Disease Biomarker: Quantitative Microstructure Measurements as Virtual Histology

Motivation: The microstructural properties of brain white matter, such as axon radius, directly affect the brain's nerve function and have been observed to correlate with various neurological diseases like multiple sclerosis and tumor-grading. Despite the importance of these microstructural properties, they have not been reliably measurable in vivo, so that their investigation has mainly relied on invasive histology examinations. These examinations in traditional neuroanatomy often introduce artifacts and cannot be used to analyze disease stages and monitor progression.

The goal here is to replace this invasive histology examination with a non-invasive in-vivo computational approach to extract specific microstructural measures of underlying tissue properties that can provide reliable and sensitive biomarkers for disease-induced changes using diffusion MRI.


Results: With my new computational approaches through analytical modeling and analysis, I was able to extract specific microstructural measures of the underlying tissue including axon sizes, distribution and volume fraction. I thus obviated some of the limitations of earlier work by (1) improving sensitivity to small axons (radius < 3 micrometer) that are more vulnerable to disease and studying the full range of axon sizes in the human brain, (2) providing more realistic model that takes into account axon radii variation, (3) making the approach applicable to low-gradient clinical scanners and whole-brain mapping by mathematically modeling the water diffusion using a new diffusion MRI protocol. My results have been presented at various peer-reviewed international conferences and my recent submission to NeuroImage is a complete validation study on the feasibility and reliability of these computational approaches. My future work in brain science will build more sophisticated computational models in order to extract other tissue compartments including neurons, glial cells and dendrites as well as microstructural properties of gray matter.



1.2 Cancer Treatment: Protein Electron Structure Calculation and Drug Design.

Another application area of my computational techniques is in drug design for cancer treatment. The method is applicable to many kinds of protein, especially cancer related proteins that the National Cancer Institute at NIH is interested in. I have been invited to present my work at NIH in 2014. Traditional drug designs are extremely costly and time-consuming: they involve a labor-intensive process of experimental screening without guiding principles from the protein-ligand electron structure. Efficient drug-discovery efforts require studying the electron structure of a protein. I have developed a unique computational drug design pipeline, which allows us to design new drugs based on the information from the electron structure and the protein properties. We currently aim to apply and refine this framework on the FDA approved drug database in order to repurpose these existing drugs for cancer treatment after radiation therapy. I will also analyze radiation-induced impact on drug target affinity.



2. Powerful visualizations for combining multidimensional scientific information

We confront a huge technological challenge in the massive data sets obtained across all scientific disciplines like biology, medical imaging and brain science. More information is being produced from various imaging modalities than the human eye can begin to make sense of. In my future research, I would continue to map these huge data sets into comprehensive visualizations that also make connections among different imaging modalities to facilitate data integration. Powerful visualization tools can significantly enhance scientific understanding in brain functions, molecular interactions, and protein network. In brain science, for example, we obtain different kinds of information from various imaging modalities such as diffusion MRI, spectral MRI, and microCT. Visualizations that integrate computationally extracted microstructural properties of white matter (e.g., axon radius and volume fraction) as well as diffusivity measures from diffusion MRI (e.g., fractional anisotropy and mean diffusivity) would greatly benefit the understanding of the brain structure and their disease-introduced changes.

Motivation: Visualize brain white-matter spatial relationships based on similarity measures from diffusion MRI for understanding brain connectivity.

Results: I demonstrated the benefits of visualization in disentangling the complex neural tractography data and in helping brain scientists understand and identify the underlying neuroanatomy and its connectivity. My ACM SIGGRAPH 2006 BEST POSTER which won FIRST PLACE in the ACM student research competition demonstrated a smooth-perceptual coloring scheme to map similar tracts to similar colors in order to help the users visually identify meaningful anatomical structures without imposing a rigid segmentation. In my later work in VIS 2007, I mapped this similarity measure into stripe texture patterns and improved the visualization of subtle differences within large white-matter structures.


Motivation: Visualize cortical-connectivity integrity and pathological changes

Results: A circular visualization interface of the tractography metrics for assessing cortical-connectivity integrity. The outer ring represents the 80 subdivided grey matter regions-of-interest (ROIs); the intensity of the edge corresponds to the normalized value of the tractography metrics across all subjects. The visualization allowed users to analyze various tractography metrics and identify the location of cortical-connectivity difference between healthy control and patient groups.

3. Interactive user interfaces for understanding and connecting data at multiple scales

Much scientific information and data are multidimensional and viewing them at different scales can drive significantly different discoveries. I would like to give scientists friendly user interfaces that can easily transition between different information scales embedded in the data. A promising approach I am interested in pursuing is to provide a multi-touch tabletop interface for the multiscale analysis of brain connectivity. In brain data, the white-matter tracts constructed from diffusion MRI provide macroscale information on the brain connectivity. At the micro-scale level, we want to examine the properties of axons, estimated using computational techniques (section 1) that make up these white-matter tracts at the macro-level. The changes observed over the course of brain diseases can differ significantly depending on the level of scale we work at. A multi-touch interface will let users transition easily between micro and macro data scales at the touch of their finger tips: navigating from axons to tensors, to white-matter tracts, and to their termination areas in the cortical grey matter. The tabletop implementation should encourage interdisciplinary and collaborative research - something that such data greatly demands. Another application domain of multi-touch interfaces I would like to explore is to provide mobile technologies that assist older adults in communication and recognition for enriching and empowering themselves in aging gracefully.

Motivation: Segmentation of white-matter pathways by interactively selecting tracts-of-interest (TOI) has become a popular way for brain scientists to test their hypotheses on white-matter connectivity and quantitative pathological analysis.

Results: I conducted evaluations and developed taxonomy and design guidelines for TOI selection as a framework for exploring and categorizing the design space of the techniques in order to analyze their utility, usability, accuracy and reliability. Good user-interfaces should be user- and data-oriented. I designed a 2D sketching interface to match the design interface to neuroscientists' training in identifying anatomical structures on 2D axis-aligned slices. The interface reduced the high-dimensional data to a 2D projection that neuroscientists can easily understand. I later extended the 2D lasso-drawing selection mechanism to 3D stereo virtual reality (VR) environment with higher-input devices for better manipulation of 3D data. I presented my results as one of the top five BEST POSTER nominees at VIS 2008. The VR environment in this interface reduced visual clutter; the higher-order input device greatly reduced navigation time (a key challenge in TOI selection tools) and especially selection time for these tortuous structures.


Interactive user interfaces also have a strong potential in enhancing student learning in schools. Two projects I am currently supervising involve understanding those effects. In the first project funded by William Beaumont School of Medicine, we are working with faculty from the Medical School in developing an anatomy learning game for first year medical students. During this development process, we are discovering the key features needed in gaming to enhance student performance in class. We are also comparing the student learning with their traditional pre-reading learning style. In the second project, we are studying the effect of mobile application on educating middle school girls about computer science. Around 800 middle school girls are being recruited to participate in this study. We hope that through this learning, we could find new ways to encourage more women to explore technology fields.