Virtual Microscopy: Distributed Image Storage, Retrieval, Analysis, and Visualization - Parallel Computing for Bioinformatics and Computational Biology - page 744

Figure 29.1 ( a ) GME using a hierarchy of authorities to manage metadata schemas. ( b ) The
flexibility of Mako database service architecture.
• Distributed storage of data sets allows parallel access to improve single user
performance.
• Distributed storage of images allows multiplexed access to improve multi-user
performance.
• Distributed location of images allows parallel image processing.
BMI developed a distributed PACS system [33], GridPACS, based on the Mobius
metadata management middleware [17, 18]. GridPACS is an application built on top
of the Mobius middleware to leverage its features. It consists of a generic image
data set schema for use with Mako to setup databases, several image ingestors for
ingesting images into the databases, and a front-end user interface for querying and
reviewing data sets. GridPACSwas designed to handle both radiology andmicroscopy
images, mainly through different ways of using the same data schema (Fig. 29.2). To
support the widest possible types of images, the schema separates the metadata from
the actual image data. Image metadata subschema allows arbitrary name-value pairs,
therefore providing a great deal of flexibility. The image data subschema organizes
data hierarchically, from image stack, to image grid, to image, and finally to tiles.
Image stack represents a sequence of images, either temporal or spatial along z -axis.
The use of tiles allows efficient storage of very large images by partitioning it into
smaller, more manageable chunks. Each tile has been labeled with the data set source,
as well as the x , y , and z extents of the bounding box. This allows efficient query and
retrieval of tiles directly given the data set id and a region of interest.
The XML schema allows Mako to rapidly create databases that are customized for
a particular application, in this case, for image management. Matching ingestors are
developed to transform source images into XML documents with the accepted syntax
and store them in the database. Two types of XML documents are submitted for each
image data set. The first is for the tiles. TheAperio ingestor imports the uncompressed
stripes, partition them into tiles, and generates an XML document for each of the tile,
attaching position and data set information. The tile XMLs are submitted to Mako,
and document references are returned and added to the data set XML document,
which is submitted to Mako after all tiles have been ingested. The actual image tiles