Page no. 30
CAD.fm
GE Healthcare
Senographe SecondLook Digital CAD System
Revision 1
Operator Manual 5189820-5-C-1EN
ROSE-1 and ROSE-S1
The first pivotal study, ROSE-1, was a multi-institutional trial to assess Second Look Analog as an aid for
radiologists in detecting breast cancer with screen-film mammography. A first supplemental study,
ROSE-S1, used digitized images obtained during the ROSE-1 study to evaluate updated software in
Second Look Analog. There were 4 components: the Missed Cancer Study assessed the percentage of
cancer cases missed by a radiologist that would be detected and worked-up with use of the system; the
Screen-Detected Study assessed the sensitivity of the system in detecting cancers on mammograms
that led to the diagnosis of breast cancer; the Reproducibility Study assessed the reproducibility of the
system’s markings; and the Normal Study assessed the false positive rate of the system.
Missed Cancer Study
The Missed Cancer Study assessed the number of previously overlooked cancers that would have been
detected and worked-up by the radiologist if using SecondLook. Seventeen (17) institutions enrolled 374
screening mammography cases that were originally interpreted as normal or benign within 9 to 24
months prior to the screening mammogram that led to cancer diagnosis. These 374 cases had both the
current and prior mammograms available for analysis. The 374 prior mammograms underwent
independent, blinded review by 3 radiologists (the panel) for detection and recommendation of work-up
of mammographic abnormalities. At least one of the panel radiologists recommended work-up in 310
cases, while the other 64 cases were not recommended for work-up by any of the panel. Of the 310
cases, 174 had one or more work-ups confirmed to be at the locations of subsequently diagnosed
cancers by 2 other (truthing) radiologists. The truthing radiologists worked independently of each other
and were required to reach a consensus over initial disagreements. They worked unblinded, with the
help of the subsequent mammogram that led to the diagnosis of cancer.
Of these 174 previously missed cancers, approximately 66% were represented primarily by "masses"
and 34% by microcalcifications. The "masses" included spiculated and non-spiculated masses,
architectural distortions, and asymmetric densities. The digitized images of these 174 mammograms
were then processed by SecondLook. The locations of the MassMarks and CalcMarks were compared
to the locations of the subsequently diagnosed cancers. This process measured the sensitivity of the
SecondLook system in detecting missed cancers. To assess the effect of the system in actual clinical
practice, it is necessary to account for the likelihood that a radiologist would indeed work-up a region
marked by SecondLook. To accomplish this, the proportion of blinded panel radiologists correctly
identifying each missed cancer case was used as a likelihood multiplier. Since there were three panel
radiologists, this proportion was 0/3, 1/3, 2/3, or 3/3. Use of this proportional weighting resulted in a
lower bound for the number of cases that showed actionable signs of cancer on the prior mammograms
that were originally interpreted as normal or benign, for the following reason. The panel radiologists who
failed to identify a region could have failed on the basis of either an error of detection or an error of
interpretation, but the distribution of cases between these two types of errors was not recorded. So it
was simply assumed that all lesions had been detected by all three of the unaided panelists and that
failures to recommend work-up were due strictly to errors of interpretation. Thus, multiplying by 0/3, 1/3,
etc. results in the most conservative estimate of the system’s effectiveness as an aid to a radiologist in
detecting breast cancer.
By this method it was determined that of these 174 missed cancer cases, 121.3 (69.7%) were
actionable. Of these actionable cases, at least 86.0 (70.9%) were marked by SecondLook and would
have been worked-up if they had been pointed out to the clinical radiologist.
