Large-scale neuroanatomical visualization using a manifold embedding approach

We present a unified framework for data processing, mining and interactive visualization of large-scale neuroanatomical databases. The input data is assumed to lie in a specific atlas space, or simply exist as a separate collection. Users can specify their own atlas for comparative analyses. The original data exist as MRI images in standard formats. It is uploaded to a remote server and processed offline by a parallelized pipeline workflow. This workflow transforms the data to represent it as both volumetric and triangular mesh cortical surfaces. We use multiresolution representations to scale complexity to data storage availability as well as graphical processing performance. Our workflow implements predefined metrics for clustering and classification, and data projection schemes to aid in visualization. Additionally the system provides a visual query interface for performing selection requests based on user-defined search criteria.