Dr. Ikeda
is an Associate Professor in the Department of Radiology at
Stanford University, Stanford, CA.
The CAD systems scan digitized film-screen mammograms,
previously loaded into special digitizers by a technologist or
technologist aide. Special computer algorithms search the digitized
film-screen mammograms for abnormal findings and mark the findings
on a low-resolution mammogram reproduction on paper or on a
monitor. The marks alert the radiologist to focus on the findings
to determine if they are possible masses or calcifications that
require further work-up, namely recall from screening. The CAD
programs are reproducible, tireless, and not distracted by the
complexities of image interpretation that influence human beings.
It was hoped that these CAD programs would aid radiologists in
detecting small cancers, which otherwise may not have been
perceived at the time of image interpretation.
This seemingly clear-cut case for using a computer to find small
cancers is a complex issue, however. Although peer-reviewed
scientific studies have shown that CAD programs can detect subtle
but suspicious mammographic findings in dense or complex breast
tissue, these studies also show that CAD programs do not detect
every breast cancer. This means that the decision to recall a
finding on a mammogram still rests solely on the radiologist's
experience and judgement in film interpretation. Secondly, the
types of CAD algorithms that search the mammograms vary from
company to company, just as two radiologists may read the same
mammogram on different receiver-operating curves, and the detection
rates for each company have never been compared. Next, the
psychological influence of the CAD program on a radiologist's
decision to recall a finding has not been studied. How does the CAD
program influence the radiologist's decision to recall a finding if
it has not been marked? What are the implications of marks on
findings that are "normal" structures? These issues regarding CAD's
role and its influence on film interpretation will certainly be
explored in greater detail over the next few years.
The other major concern regarding implementing CAD programs is
its cost. Issues of reimbursement in this economically challenged
medical environment influence whether institutions will purchase
CAD equipment and whether the cost of running this program will be
justified. From an economic point of view, the cost of the CAD
equipment and personnel salaries to digitize the films and collate
the CAD output with the screening mammograms all should be
considered. Does purchase of a CAD program make economic sense?
This supplement covers these issues and explores new territories
in which CAD is being used to evaluate mammograms in clinical
practice. The articles review the scientific literature relating to
methods of obtaining CAD output and to CAD's use in clinical
practice for screening mammograms. The authors raise issues of
reimbursement and examine the advantages and disadvantages of using
CAD programs. These articles also answer the difficult questions of
"What is CAD?" "How does CAD work?" and "What has the scientific
literature shown regarding CAD and screening mammograms?" After
reading this suppement, a radiologist may have a more informed
answer to the important question, "Is CAD right for my practice
right now?"