Biomedical Engineering Reference
In-Depth Information
language with access to sockets and HTTP protocols. In several cases, there exists
a language API to access all facilities within Bisque (initially MATLAB, Python,
and Java, but the list will grow). However, since all services are accessible through
HTTP protocols, any client program with Web access will be able to interact with
our system.
Two styles of analysis modules are available. The first is termed ''internal''
for an analysis available through the Web interface and which uses our parallel
execution engine. These are commonly used for analyses that have been found
useful to a large number of users or that need significant computational resources.
We also support an ''external'' analysis for those that need to operate outside
the core Bisque framework. In essence, these analyses are external programs that
utilize Bisque resources. It is often easier to initially create an external analysis and
migrate it to an internal module as needed.
Both styles of analysis modules have access to the same Bisque services. In
particular, analysis modules may create DoughDB objects, tags, and links through
the data service, execute other modules, and store new images and masks through
the image service. This key feature promises to make integrating analysis modules
much simpler than current methods.
13.4 Analysis Architectures for Future Applications
Image management systems of the future will need to provide additional capa-
bilities such as supporting the use of external computing resources for extensive
computation-bound tasks and the use of libraries for modeling of complex systems.
In this section, we develop the idea of probabilistic data management and analysis,
which is becoming increasingly important for image databases.
The primary methods of scientific experimentation are becoming high through-
put, producing vast amounts of data. However, much of the data being produced
is not in the form of definitive measurements, but rather contains an inherent
uncertainty from the acquisition process. The current state of scientific analysis
does not usually take this uncertainty into account. In microscopy, for example,
poor image acquisition, light pollution, and experimental error all produce images
with poorly demarcated cell or tissue boundaries. The typical automated process
of analysis on these images performs a threshold-based segmentation followed by
feature extraction. While such analyses can provide accurate results, they are usu-
ally limited to high-quality, high-resolution images and only present one possible
interpretation of the data [37]. No notion of uncertainty is presented; thus some
of the original data is, in essence, lost.
Elements of uncertainty arise in all stages of the analysis work-flow includ-
ing image acquisition, feature extraction, querying and management, and pattern
discovery. Aspects of uncertainty must be explicitly modeled and accounted for
in all stages, and their precise impact understood in a holistic manner. While
the main context of this chapter is biological images, the ideas extend to a va-
riety of applications in other image-based domains, such as environmental man-
agement, geographical information science, remote sensing, and interactive digital
multimedia.
Search WWH ::
Custom Search