This poster is published under an
open license. Please read the
disclaimer for further details.
Keywords:
Breast, Mammography, Technology assessment, Cancer
Authors:
C. Tromans1, R. Highnam1, O. Morrish2, R. Black2, L. Tucker2, F. J. Gilbert2; 1Wellington/NZ, 2Cambridge/UK
DOI:
10.1594/ecr2014/C-0363
Results
Examples of some of the different image types that could potentially be used by radiologists to visually assess breast density are given in Figure 1 (note: as the C-View module is optional,
synthetic CC and MLO views were not available for the current study).
A comparison of the means for the various Volpara parameters,
indicate that there was no significant difference between VBD,
BV and FGV estimates from conventional mammograms or the DBT projections (Figure 2).
The overall mean VBD as assessed from the DBT images was lower than the VBD from the conventional mammograms,
but this change was not statistically significant.
The breast volumes as estimated from the conventional mammograms and the raw DBT projections were near identical,
with a PCC of 0.998.
As seen in Figure 3,
variation was highest for the largest breast volumes.
The VBD and FGV estimates were also highly correlated between the 2D images and DBT projections,
with PCCs of 0.903 and 0.919,
respectively (see Figures 4 and 5).
The Volpara generated BI-RADS density scores matched for 19 out of 20 cases,
with a weighted kappa score of κ = 0.953,
indicating that there was very good agreement between density assessed from the DBT projections and the conventional mammograms (Figure 6).
It was noted,
however,
that Volpara outputs for very dense breasts appeared to be lower on the DBT projections compared to the conventional mammograms,
and reasons for this are currently under investigation.