Hope in the Fight Against Breast Cancer: Cancer Detection
Almost 179,000 American women will be diagnosed with
invasive breast cancer in 1998, according to American Cancer Society statistics.
During that same period, more than 43,000 women will die from the disease.
Since their introduction in the 1960s, mammograms have become a powerful
weapon in the fight against breast cancer. When used by experienced radiologists,
the special x-ray technique can catch cancer during its early stages, when
it is most treatable. Yet, despite the advances in cancer detection made
possible by mammography, current methods have their limitations.
With funding from NSF and other agencies, hope is on the horizon. Digital
mammography, a new technology now being evaluated by the Food and Drug Administration
(FDA), promises to locate very small tumors using lower doses of radiation.
In the June 1992 issue of Radiology, Dr. Faina Shtern of the National
Cancer Institute wrote that digital mammography is "the most fertile territory
for major advances in the detection and diagnosis of minimal breast cancers."
A VOICE IN THE WILDERNESS
Robert Gray, professor of electrical engineering at Stanford University,
and Richard Olshen, professor of health research and policy, are two of
the people helping to create these advances. Gray, Olshen and their colleagues
have developed a method for comparing the quality and precision of traditional
mammograms to their new digital counterparts. A primary goal of this portion
of Gray and Olshen's NSF-funded research has been to develop standards for
evaluating the reliability of the digital images.
In digital mammography, the analog film normally used for recording the
image is replaced by an electronic x-ray detector. The signal from the detector
is digitized and stored in the computer's memory, and the resulting image
can be displayed and adjusted by the radiologist. This facilitates detection
of small breast cancers and microcalcifications, or benign calcium deposits
in breast tissue.
Gray and Olshen began working on a way to reliably compare the accuracy
of traditional and digital mammograms in the early 1990s. They became interested
in signal processing and data compression for medical imaging, in part because
of concerns expressed by some physicians that the processes for acquiring,
storing and transmitting digital images would result in a loss of reliability.
Before the new digital techniques can be used in clinical settings, physicians
and the FDA must be convinced that digital mammography is as accurate as
its conventional counterpart. Toward that end, the Stanford team, led by
Gray and Olshen and with additional support from the U.S. Army, has used
its own algorithms to compare more than 500 digital and conventional mammograms.
Some of their results have already been incorporated into the FDA's new
protocol for evaluating digital mammography systems. Several equipment manufacturers
are testing these systems; one company submitted data from clinical trials
to the FDA in late 1997. The data from the Stanford research will help the
FDA evaluate the systems and determine whether or not they should be marketed.
"The newer (digital) machines filter the signal to eliminate the noise,
clean up the image and deliver better resolution," explains John Cozzens,
program manager in NSF's Directorate for Computer and Information Science
and Engineering. "Radiology is not an exact science," he continues, "which
to-date has made it incredibly difficult to confirm the accuracy of digital
mammography. When Bob first began his work, he was a voice in the wilderness."
According to Cozzens, the thrust of the work is to demonstrate that there
is no difference between conventional and digital mammography and, in fact,
that the digital technology will prove to be better. "Doctors need to be
confident that the digital images are as good as the films they are used
to reading," says Gray. "There is mounting evidence that digital mammography
will be superior to analog film methods."
Many researchers believe this new imaging technique may provide significant
benefits over traditional methods.
Unlike film, which must be developed and yields one copy, digital mammograms
can be duplicated and transmitted over the Internet to multiple locations.
Once approved by the FDA for commercial use, the technology will enable
physicians or technicians in remote areas to employ telemammography--sending
mammograms over the Internet--to confer with specialists at large medical
centers for initial readings or second opinions.
Digital mammograms can easily be compiled into a database and be used
as a powerful teaching tool. The image files will enable medical students,
interns and radiologists to study, classify and compare hundreds of cases.
In addition, the digital approach delivers a lower dose of radiation and
may improve imaging of particularly dense breast tissue.
As computers enter the realm of mammography, they also have the potential
to assist radiologists in evaluating what they see. With a grant from NSF,
Professor Ed Delp and his colleagues at Purdue University in Indiana began
work on a computer-aided diagnosis (CAD) system in 1992. His research involved
the development of computer algorithms that recognize and highlight irregularities
In this foundational research, Delp set out to look for stellate lesions.
Once he had identified the characteristics of these star-shaped tumors,
the research team had planned to design an algorithm that would flag possible
lesions, to assist in the reading of the mammogram.
In the process, however, Delp discovered that many physicians were resistant
to the idea of a CAD system for identifying tumors. Based on that reaction,
Delp decided it made sense, instead, to look for "normalities" in
breast tissue. "When you've defined what is normal, you know that everything
else is abnormal," explains Delp.
On a related project, funded by the National Institutes of Health (NIH),
Delp's group pioneered other techniques such as multi-resolution processing
and tree classification. By examining the digital images at many different
resolutions, researchers are better able to classify the information contained
in the mammograms. Tree classification is a method for extracting the unique
features of each image and then classifying it as either normal or abnormal.
"The preliminary findings in our search for 'normals' are promising," Delp
notes. "This new kind of CAD has enormous potential. It will save time
by allowing physicians to concentrate only on the mammograms that show abnormalities." Delp
is planning a rigorous three-year study to evaluate the effectiveness of
his CAD system.
Dragana Brzakovic, an NSF program director and expert on digital mammography,
sums up the anticipated advantages of the research that she, Delp and others
are conducting: "With digital mammography, you can control quality
in real time. Radiologists will be able to evaluate mammograms on the spot--while
the patient is still there--and correct positioning and other problems immediately." She
adds, "Women will no longer have to live in fear, waiting days for
THE SCIENCE OF HOPE
Concern about early cancer detection is widespread. The work of Gray, Olshen,
Delp and other NSF-funded scientists has paved the way for other research
in the field of digital mammography.
- Researchers at the National Aeronautics and Space Administration's
Langley Research Center have teamed with colleagues at the University
of Virginia and University of Massachusetts Schools of Medicine
to test concepts that make all-digital full-breast imaging possible.
The first working system was introduced in November 1995 and
is now being tested at four locations in the United States and
- At the Naval Surface Warfare Center, researchers are beginning
to transfer imaging techniques developed for the Department of
Defense to the development of their version of a CAD system.
- With a grant from the U.S. Army, scientists at the University
of South Florida have established the Digital Database for Screening
Mammography to facilitate sound research into the development
of computer algorithms to aid in screening.
- The effort to improve breast cancer detection crosses many
borders. Among those working to improve the technology and techniques
are researchers at Brandeis University and Lawrence Livermore
National Laboratory, as well as at England's Oxford University
and Norway's Signal Processing Group.
The work in this field is bringing computer scientists, engineers and statisticians
together with physicians and patients from around the world. They are working
together to find better ways to identify and treat breast cancer, bringing
hope to tens of thousands of women and their families.
THE FRENCH CONNECTION (SIDEBAR)
Under an NSF grant awarded in April 1997, Robert
Gray, Stanford University professor of electrical engineering, has been
collaborating with mathematicians at the French National Center for Scientific
Research (Le Centre National de la Recherche Scientifique, CNRS). CNRS
is the largest research organization in France for fundamental research
in science and engineering. This international team is designing algorithms
for image compression and classification. Their work is expected to contribute
significantly to further advances in digital mammography.
"This collaboration brings the value of France's
sophisticated theoretical mathematicians to bear on Gray's work," says
Rose Gombay, program manager in NSF's Division of International Programs. "Grants
from NSF's Division of International Programs are designed to stimulate
international collaborations and advance U.S. science. Our goal is to
provide catalytic funding so that U.S. researchers will incorporate such
collaborations in their plans," she concludes.