Biomedical Engineering Reference
In-Depth Information
processing. Methods to process unstructured images, extract feature descriptions
that can represent the content effectively, and organize these descriptions into
a searchable database are needed and are of immense value to the scientific
community. These new methods enable biologists to look for patterns in ways
that are not currently possible, and use emerging data mining tools to discover
new patterns that enhance our understanding of the various biological events
at the cellular and subcellular level. Scientists are able to search these image
databases by image keywords like texture patterns or time-dynamics information.
They can browse, navigate, and explore such databases in a manner that is fast,
intuitive, and most likely to reveal connections among images.
There has been significant progress in computer vision and database research
dealing with large amounts of images and video. In computer vision, content
based image retrieval (CBIR) has been an active research area for over a decade.
The CBIR paradigm is to represent the visual content in images using low level
image descriptors such as shape, texture, color, and motion, as applicable to a
particular application domain.
13.1.3 State of the Art: PSLID, OME, and OMERO
Several large projects have paved the way for large scale image analysis. One of the
earliest systems developed was the Protein Subcellular Localization Image Database
(PSLID) [16] which provides both management of datasets and analysis. Its fea-
tures include searching for images by context (annotations) or content, ranking
images by typicality within a set (e.g., to choose an image for presentation or pub-
lication), ranking images by similarity to one or more query images ''searching by
image content'' or ''relevance feedback,'' comparing two sets of images (hypoth-
esis testing) (e.g., to determine whether a drug alters the distribution of a tagged
protein), training a classifier to recognize subcellular patterns, and using a trained
classifier to assign images to pattern classes (e.g., assigning images to ''positive'' or
''negative''). The project provides datasets and Java-based software for analysis.
Its functionality is available through a Web browser.
The Open Microscopy Environment (OME) [14] has had considerable success.
Its long feature list includes support for large datasets, integrated analysis system,
and support for data provenance. The original OME server was written in Perl
and supports MATLAB integration. It provides access to clients through a Web
browser or custom programs (VisBio).
The OMERO [15], also part of the umbrella OME project, is a Java-based
system for managing, annotating, and visualizing images and metadata. It is
formed out of two main subprojects: OMERO.server, and OMERO.insight. The
OMERO.server is a Java-based database server, while OMERO.insight is a client
application for organizing, viewing, and annotating images.
13.2 Rationale for Bisque
We review the motivation for Bisque, an image management and analysis system
being developed at the Center for BioImage Informatics, University of California
Search WWH ::
Custom Search