Biological Imaging

Understanding the genotype-phenotype relationship is one of the most ambitious tasks for biomedical researchers in the post-genomic era. By characterizing different phenotypes associated with different genotypes for an organism, researchers are able to obtain profound information about the roles and functions of the genes. Nevertheless, accurate characterization of the phenotypes is itself a formidable challenge. For an organism, its phenotype is constituted by its “appearances” (e.g., morphology and structure) at various scales ranging from molecular, cellular and tissue levels to physiological and behavioral levels. Currently many phenotyping tools are focused either on the molecular level (e.g., microarray) or on the physiological and behavioral levels. Notably, there is a lack of systematic means for characterizing the phenotypes at cellular and tissue levels even though the cellular and tissue morphology and structures offer critical information about the mechanisms of important biological processes such as how cancer initiates in the tumor microenvironment. The major reason for this gap is the challenging problem of delineating the intricate cellular or tissue structures in both 2D and 3D spaces with spatial scales spanning several magnitudes (i.e., from nanometer to centimeter). Imaging techniques are usually used to study the structure and morphology of cells and tissues. However, any specific imaging modality such as light microscopy or ultrasound imaging can only cover a specific range of spatial resolutions.

Our group takes a system biology approach for the development of a phenotyping system using multiresolution and multimodal imaging techniques. In order to obtain a comprehensive characterization of the cellular and tissue phenotypes, researchers have to use multiple modalities of imaging technologies for multiple resolutions. For instance, microscopic imaging is largely used to investigate the cellular structures at micron resolution while high frequency ultrasound imaging is used to characterize tissue features at millimeter scale. In addition, different imaging modalities also offer different types of information ranging from structural to functional.

During the past two years, our group has developed a series of quantification tools for analyzing large set of biomedical images including microscopy images, MRI images, high frequency ultrasound images, and optical coherent tomography images. These tools include novel algorithms for image registration, 3D reconstruction, image segmentation, and visualization of the large set of images in the scale of terabytes. Specifically, we have developed a two-stage automatic deformable registration algorithm based on matching conspicuous anatomical structures between consecutive images. This algorithm is highly parallelizable and has been applied for processing large set of images. The segmentation algorithm is based on the notion that different tissues are different types of heterogeneous biomaterials and hence can be segmented using the spatial statistical features (two-point correlation function) which are used to characterize such materials. The segmentation algorithm also formed the basis for developing the probability-based transfer function for the volumetric rendering technique in the visualization process. These algorithms and the processing pipeline have been applied in several biomedical applications including investigation on the role of Rb, a well-known tumor suppressor gene, in mouse placental development.

On the application side, we focus on applying our system to solve high-impact biomedical problems. We are currently collaborating with Drs. Gustavo Leone and Michael Ostrowski in the OSU Comprehensive Cancer Center to study the phenotypes of the tumor microenvironment. Specially, we plan to use the above mentioned techniques to develop high resolution cellular and molecular atlas for the tumor microenvironment at various tumor progression stages. These atlases will be critical for developing and evaluating new chemotherapeutical agents for cancer treatment.

As the outcome of the project, the phenotyping system will allow the researcher to investigate the complicated structures of biological samples with unprecedented accuracy and efficiency. The capacity for quantitatively characterizing the phenotypes at full scales will significantly promote the understanding of the roles and functions of various genes and the mechanisms underlying important biological processes, which will tremendously benefit the development of new treatment methods for diseases such as cancers. In engineering, the theory and algorithms as well as the high performance computing systems developed to cope with the computing challenges will widely impact on the progress in areas such as computer vision, scientific visualization, machine learning, supercomputing, and software architecture. In addition, the developed phenotyping system will be of great importance to many other areas such as material sciences and nanosciences which require visualization and quantification of structures at high resolution.

Project Researchers

Kun Huang, Ph.D.
Tony Pan, M.S.
Ashish Sharma, Ph.D.

Project Publications