Abstract:
HTML
Purpose: To evaluate the performance of a segmentation algorithm to
automatically remove bone structures from CT angiography data to enable
vascular visualization.
Methods and Materials: Eight contiguous head CTA data sets, acquired using
different configurations of CT scanners (GE Medical Systems, Milwaukee, USA),
were retrospectively analyzed in this IRB-approved study. The data sets, which
contained a variety of pathologies,were anonymized for patient privacy and
transferred to a PC workstation in DICOM format. An automatic partition
algorithm identified a partition line superior to the Circle of Willis, as
indicated by the vessels not being close to bone in the corresponding axial
slice. The top and bottom partitions were processed independently and were
subsequently merged using a constrained region growing to correct
discontinuities. The segmented bone regions were masked on the original volume
to suppress the bone voxels. The vascular trees were volume rendered(VR) by a
single user both in 3-D and using the Maximum Intensity Projection (MIP)
method. Stationary images were created for the following standard views: AP,
left and right lateral, superior (top), and oblique (VR only). For each view,
the VR and MIP images were created and placed side by side in a viewing window.
A radiologist specializing in neuro-CT reviewed the prepared 3-D and MIP views
for each case and responded to a prepared set of questions for rating the
algorithm performance. The rating was based on a scale of 1-5, where 1 represented
low performance and 5 represented high performance. The questions were based on
several anatomical features potentially affected by the removal of bone
surrounding the major vascular structures. The rating was analyzed to evaluate
the strengths and weaknesses of the algorithm.
Results: The mean rating (1: Reject, 5: Accept) of this bone segmentation
algorithm was 2.54, (Std. Dev.: 0.64). The algorithm was rated highest for the
carotids merging into the Circle of Willis and for the clarity in the visualization
of inter-cranial aneurysms. The algorithm received the lowest rating for the
cavernous-petrous segment of the internal carotids in the skull base region and
in removing the bone cleanly in this region.
Conclusion: A bone segmentation algorithm developed for head CT angiography
applications was evaluated on 8 cases. This study forms the baseline for our
continuing efforts to automate the process of vascular visualization in a
complex anatomical region with various pathological conditions and expanding
this application into neck CTA. (S.S., R.M. and Y.M. are employees of GE
Corporation; C.P.W., C.M. and K.T. have a financial relationship.)
Questions about this event email: srikanth.suryanarayanan@ge.com
Suryanarayanan PhD, S,
Automatic Bone Removal for Head CTA: A Preliminary Review. Radiological Society of North America 2003 Scientific Assembly and Annual Meeting, November 30 - December 5, 2003 ,Chicago IL.
http://archive.rsna.org/2003/3107556.html
