Improvement of Registration of
2D X-ray Mammograms and 3D Speed of Sound Images
Julie Bonn1, Torsten Hopp,
Nicole V. Ruiter1, Neb Duric
ABSTRACT: Breast cancer is the most common cancer in
women in the US.
It is known that early detection is key to survival rates. Currently, the
standard screening method, X-ray mammography, has limited tumor visibility in
dense breasts as it results in projection images of the deformed breast. A new
imaging modality, Ultrasound Computer Tomography (USCT), generates 3D images of
the non-deformed breast. Since USCT is still in development, a registration of
the images from the two modalities is of interest to analyze the accuracy and
quality of USCT. Previously, we presented a method of registration of
mammograms and USCT speed of sound images; based on that, a larger dataset was
registered in this work. The data files were received pre-segmented. Due to our
need for a closely cropped and smoothed image and complete circumference
alignment between the two modalities, manual corrections to the images were
necessary: freehand corrections to the pre-segmented USCT images and manual
angle of rotation setting due to different patient positioning. The preliminary
accuracy of registration was improved by almost 10 mm per patient. For a total
of 15 datasets, the final results showed a mean center of lesion displacement
of 7.1 mm (SD = 5.4) and a mean
lesion overlap of 91% between the two imaging modalities. This illustrates a
high level of accuracy, affirming previous results and leading towards clinical
applicability. Current research is focused on fully automating the
registration.
Barbara Ann
Karmanos Cancer Institute, Department of Radiation Oncology, Detroit, USA