CHICAGOTwo studies presented at the 87th Scientific Assembly and Annual
Meeting of the Radiological Society of North America (RSNA) indicate that
computer-aided detection (CAD) of lesions on conventional mammography studies
is comparable to double-read mammograms, and it produces fewer false-negative
Robert A. Schmidt, MD, associate professor of radiology, New York University
Medical School, assessed the ability of CAD to find lesions radiologists
commonly overlook. In this study (abstract 1001), radiologists participating in
a continuing medical education program as well as three expert readers were
asked to interpret copies of 100 screening mammograms.
The films contained 50 biopsy-proven cancers in 45 patients and 55 benign or
normal cases. Only 49 original mammograms were available. Both those 49
original films and the 100 copies were digitized and analyzed by version 2.2 of
the ImageChecker (R2 Technology, Los Altos, California).
Radiologists were told to grade mammograms as normal or abnormal, then
characterize every abnormal lesion that was BI-RADS 3 or higher. A mark by a
radiologist was labeled correct if it was within a distance of no more than one
third of the breast dimension from the actual location of a lesion.
"We were much more stringent on the computer marks," Dr. Schmidt
said. A mark by the CAD system that was within 1 lesion width or less of an
abnormality was considered to be positive, and CAD had to mark the correct
lesion type (mass or calcification).
There was wide variability among the radiologists in the accuracy of their
interpretations. "Some radiologists were relatively low in performance,
and some beginners were better than experts," Dr. Schmidt observed.
Overall, the general radiologists detected 70% of the cancers, and the
experts detected 81%. The CAD system marked 80% of the cancers in both original
and copy films, Dr. Schmidt said.