|
|
Like many clinics
of this type, researchers there rely heavily on ultrasound technology
to generate pictures of internal body tissues. But Roger Pierson, an associate
professor and the unit's director, has developed a method to enhance those
images, opening new areas of medical inquiry. The System Over the past five
years, Pierson has put together a computer-based system in which preprocessed
ultrasound images are acquired by a digital acquisition system as they
stream off the digital scan connector within the ultrasound device. They
are then ported to a powerful SunTM Microsystems workstation
for interrogation with what has come to be called visual data analysis
(VDA) software. "We grab the data just before it goes into the video
processor and stream it into the computers," Pierson says. To analyze the data,
Pierson and his team use PV-WAVE®, a VDA package from Visual
Numerics, Inc. of Houston, Texas. PV-WAVE includes a fourth-generation
language (4GL) for creating custom analysis and visualization routines,
along with a library of canned functions for such tasks as convolution,
filtering, and edge enhancement. Advanced math functions include Gaussian
integrals, Fast Fourier Transforms (FFT), data point differentiation and
interpolation. "The software
does some things exceptionally well, such as intensity mapping over images,"
Pierson says. "You can flip up an image and make a diagnosis very
quickly. It helps keep track of all the images and provides the building
blocks for our specialized research and analysis." John Deptuch, a computer
programmer who works with Pierson in the lab, says the most useful routine
is one that allows him to take a two-dimensional array and shade it as
a three-dimensional surface, where the height is based on the value in
the array. "Writing that routine myself would have been an awful
lot of work," Deptuch says. The Application Researchers at the
University's College of Medicine are using the system in several applications.
By studying computerized images of ovarian follicles, for example, they
are learning to tell good follicles from bad. Good follicles contain a
visible egg and are likely to perform their proper physiologic function.
This increases the success rates for conception. In the long run, the
information gathered could help scientists understand how to "turn
on" the egg-producing mechanism in cases in which it is not functioning
properly. Alternatively, researchers also want to understand how to turn
the egg-producing mechanism off for contraceptive purposes. Another very
promising area is cancer research. When Pierson and his team turned their
equipment to study ovarian tumors, they realized that the visual data
analysis could help them diagnose whether a cyst or tumor is malignant
or benign based on a computer analysis of its structure. This insight quickly
spread to an interdisciplinary research project to verify the findings.
Herb Yang in the university's Computer Science Department works closely
with Pierson to guide the research. Today the team includes people throughout
the university and medical community, from surgeons to computer graphics
specialists, working together to establish a rigorous scientific and statistical
basis for expanding these important findings beyond the clinic. Old and New Analysis Standard output from
an ultrasound machine produces black-and-white images that are studied
against a light board or on a video monitor. Whereas most people can only
discern about 80 shades of gray through a visual inspection of ultrasound
images, Pierson's computer can distinguish 256 shades. The addition of
bandwidth filters, superimposed colors, and 3-D visualization techniques
from PV-WAVE enables even finer distinctions and variations to be observed. "For example,
during a metastatic process, the blood flow to the organ increases,"
Pierson explains. This becomes apparent in an ultrasound image because
the soundwaves decrease in their intensity and amplitude. Blood is a fluid,
and fluids reflect a lower value echo. But in contrast to standard approaches,
the computer can discern subtle differences that would be difficult to
detect, such as the difference between an amplitude of 130 and an amplitude
of 150 from a visual perusal of a black-and-white photographic image. "Instead of
an arbitrary scale of reference, we have exact figures that we can apply
against an absolute scale," Pierson observes. The distinctions
Pierson can now make are important in understanding the processes associated
with infertility because the release of an egg from the follicle is related
to blood flow and to the action of the hormones on the tissue of the follicle.
For example, when the leutinizing hormone (LH) is present in sufficient
quantities to trigger the release of an egg, it causes those cells to
hypertrophy. This gives off a different echo-intensity pattern than the
pattern received from cells that don't respond to that hormone. "We can look
at the tissue and determine LH response based on tissue response,"
Pierson says. "If the LH receptors aren't there, the follicle tends
to be very thin and bright-walled. The computer allows us to put hard
numbers on these subtle changes. From these we can make predictions, such
as, within this range of numbers, the likelihood of releasing an egg is
80 percent." The new techniques
are equally useful in the study of contraception. "We're trying to
understand what the hormones released by the pituitary gland do to the
ovaries on a monthly basis," Pierson explains. "We can use our
infertility patients as models of the hormonal state necessary to prevent
conception." Buoyed by their results
in infertility and contraception studies, Pierson and his colleagues have
now turned to the study and detection of breast tumors to establish whether
they can detect lumps that are not apparent through a physical exam, whether
they can tell a fast-growing lump from a slow-growing one by its visual
patterns, and, ultimately, a fibrocystic lump from a malignancy. "Based on what
we have seen so far, there is a different visual pattern displayed in
the benign lumps than there is in the malignant ones, but it is too early
to say much more," Pierson says. "We need to be able to identify
those different tissue types, and we need to get the segmentation of the
image to the point where we can reliably detect the lumps. Then we can
begin to apply that knowledge to a generalized screening scan." Looking Ahead It's taken Pierson and his team about five years to perfect the techniques of ultrasound visualization. "Its worked; that's been the thrilling part," Pierson says. "Once you understand the physical components of bouncing sound through someone's body and generating an image from that, everything you do after that is nothing more than standard image-processing," Pierson explains. "It's what you interpret from the results of your studies that's important, and that is what the VDA software has given us -- a new way to interpret our data and better ways to manipulate it." |
