GEMIAC has initiated several collaborative efforts with the Biomedical Informatics Research Network BIRN). A small 3-node BIRN cluster has been set up at the Department of Biomedical Informatics (BMI), in order to serve as a BIRN storage and computation node. In a joint work, we have developed a prototype system, using our DataCutter middleware, to support querying and processing of large microscopy images stored in BIRN repositories accessible through Storage Resource Broker (SRB).
The prototype allows a researcher to query microscopy datasets stored in SRB through the Telescience portal, extract the subset of the microscopy images as requested by the query, and process them on a cluster of PCs. The project involves the use of the Telescience portal to formulate a client query, the DataCutter and SRB systems to retrieve the subsets of the microscopy images from repositories at SDSC, and DataCutter to process the data on a cluster of PCs at OSU.
We are also collaborating with Dr. Maryann Martone of UCSD NCMIR on 2D reconstruction of confocal images for mouse brain studies. Mouse brain offers a convenient model for studying neuroscience. The goal of the mouse branch of BIRN is to create an integrated anatomical atlas, which provides spatial correlation for datasets ranging from CT, MR, bright-field microscopy, confocal microscopy, as well as molecular data such as genomic and protenomic data. The atlas also serves as a visual query system.
One of the challenges in creating and integrating such an atlas is in the amount of data one must work with. Our interaction with the BIRN group specifically deals with data from their 2-photon confocal microscope. The microscope generates data as tiled images that are then stitched as a three-dimensional stack of image mosaics. While the individual image tiles are small (512 by 480 pixels), a whole image for a single z-plane is far larger. Current capability of the scanner and the amount of data it generates has already exceeded the capability of the existing software tool for stitching the images together.
The goal of this collaboration is to create distributed tools that allow parallel image stitching and volumetric visualization operations to occur in a Grid environment. Java-based software has been developed to support the 2D image reconstruction in a cross platform fashion. The goal is both to enhance image stitching performance as well as enable large image processing.
For visualization and 2D rendering of the stitched images, Aperio's free ImageScope software is used. This software reads a collection of stripes as TIFF images and visualizes the stripes as a zoomable and pannable image. Initial results show that the naïve stitching approach, with 10% overlap, is not sufficient. There are noticeable seam artifacts in the resultant imageset.
We are working on adapting BIRN's existing software for distributed computation. There is also a significant level of background intensity variation due to optics. Current work includes adaptation of advanced and adaptive image stitching algorithms for this application in a distributed environment. The images also require background correction, which will facilitate the computer analysis downstream.