Biology
University of Washington
Building blocks to better medicine-creating a structural biology knowledge base
Reproduced with permission from NeXT
Computer, Inc.
A Reference
Guide to NeXT in Higher Education, Fall 1992
ยช
1992 NeXT Computer, Inc
A comprehensive understanding of structural biology is key
to the practice of medicine. At the University of
Washington, James Brinkley and Bill Barker of the
Structural Informatics Group are using a network of NeXT
computers to create a knowledge base that allows medical
professionals to more easily access information related to
structural biology.
Because the amount of structural information needed to
create a knowledge base of human anatomy is too great for
one research group to gather, the university's team is
designing a framework to allow other research groups to
build discrete portions of this knowledge base. "We chose
NeXTSTEP as our primary knowledge base development platform
because of its outstanding object-oriented programming
environment and the ease with which we'll be able to build
distributed, object-oriented knowledge-base systems to test
our ideas," explains Brinkley.
Via computer networks, the modules can run on different
computers in scattered locations worldwide. "The design is
that of a distributed system consisting of many
independently developed modules running either on the same
or different computers," says Brinkley.
Since the problem of representing structural knowledge is
far from solved, Brinkley's group is concentrating on
specific problems-the incremental solution of which will
lead to better methods for representing that knowledge.
One such problem is finding specific biological objects in
medical images. This occurs in many areas of medicine-from
identification of cells on microscopic slides to extraction
of organs in 3D image data. For example, physicians need to
extract the kidney from computed tomography images of
patients undergoing radiation treatment for cancer. Once
the kidney is isolated, other computer programs can plan
the treatment so minimal radiation reaches the kidney and
maximal radiation reaches the tumor. Currently, experts
trained in anatomy manually trace these structures in
medical images. Because manual tracing takes a tremendous
amount of time, Brinkley wrote an application that uses
knowledge of anatomy to partially automate this process.
The application, SCANNER, takes a cross-sectional image of
part of the human body and isolates an organ such as a
kidney with an image. With an initial starting point
entered by the user, SCANNER retrieves a stored model
representing spatial knowledge of the expected shape and
range of variation for the class of all kidneys. This model
then defines a region on the image within which the
computer searches for the kidney borders. The model also
defines a current best guess as to the countour of the
kidney. As kidney borders are found, the search region
shrinks and the best guess contour is adjusted until it
matches the shape of the kidney in the image.
As an offshoot of the SCANNER project, the group created an
object-oriented medical image-processing framework that
works as a frontend to other image processors. Brinkley
credits the NeXT machine's distributed object-oriented
environment for making this task relatively simple.
"As our knowledge bases evolve and as we work with other
groups, we expect the NeXT computer will continue to play a
major role as an intelligent center of our knowledge base
systems. We have increased our ability to obtain
information three-fold thanks to the NeXT machine."
For more information, please contact:
James Brinkley
Research Assistant Professor
University of Washington
Seattle, WA 98195
(206) 543-3954
brinkley@synapse.biostr.washington.edu