In the work of Zanca
et al.,11 the entire calcification clusters were extracted in a
single operation. This has the advantage of maintaining the
spatial orientation of the calcifications within the cluster,
which can be a predictive factor of malignancy. However,
the anatomical background within the cluster can generate
undesirable structures surrounding the calcifications. Therefore, in this work, each calcification was extracted separately. A small region of interest (ROIcalc) containing each
calcification was cropped from the image of the breast tissue
sample. The mean background pixel value within ROIcalc
was calculated from the mean pixel value of smaller regions
drawn within ROIcalc but outside of the calcification. Each
pixel value in ROIcalc was divided by the mean background
pixel value, generating a normalized calcification image.
The pixels in ROIcalc with values higher than the mean background (due to the nonuniform soft tissue structure) resulted
in pixel values greater than one in the normalized calcification image.