In this project, we will develop a commercial resource for the automated analysis of brain anatomy, based on MRI. This product is based on the whole-brain parcellation algorithm with the following unique features. First, it is based on a cutting-edge multi-atlas approach, in which we will incorporate rich atlas resources from Dr. Mori's lab at the Johns Hopkins University (JHU). Second, our multi-atlas approach is based on advanced diffeomorphic image transformation and multi-atlas probability fusion, recently developed by Dr. Miller at JHU. These CPU-intensive algorithms, combined with a large atlas inventory, require highly parallelized computational resources. We, therefore, will develop a fully portable and scalable cloud-based architecture, such that many users can have access at minimum costs. Third, we will develop a flexible architecture to define brain structures with multiple anatomical criteria, providing a very unique multi-granularity analysis, which provides an anatomy-centric and intuitive interface for clinical use. Fourth, we extend the analysis to diffusion tensor imaging (DTI) by incorporating a unique approach to multi-contrast image transformation and probability fusion. Last but not least, these algorithms can convert a set of multiple MR images to a quantitative and standardized Anatomical Matrix, which allows us to perform image data structurization, searching, and individualized analysis of anatomical phenotypes.
Aim 1 : To establish a cloud-based servicing architecture: We will develop a scalable and portable architecture for cloud-based computation. Parallel processing is required to achieve fast computation for the multi-atlas calculations. The algorithms accept DICOM data from four major vendors and apply a parcellation tool that identifies 254 brain structures.
Aim 2 : To establish a web-based interface for non-corporate users: To make our advanced image analysis tools widely available for research communities, we will create a web-based interface and provide the service at a minimum cost ($20/data).
Aim 3 : To implement a data visualization interface with ontology-based multi-granularity analysis: Our image analysis pipeline is a departure from conventional voxel-based automated analysis. Our structure-based analysis reduces the anatomical dimension to much lower scales. However, there are multiple ways to perform the structure-based information reduction. The ontology-based analysis provides a novel way to perform hierarchical anatomical interpretation of the structure-based analysis.
Aim 4 : To increase the number of atlases and cases in the database for interpretation support: Through the collaboration with JHU, we have access to a large inventory of research and clinical data, including controls and various patient groups. To create reference data, we will process these data and establish a background database, against which users can compare and interpret their data.