[Special Date] Iterative 3D/4D Cone Beam CT Reconstruction on GPU in Cancer Radiation Therapy

Applied and Computational Mathematics Seminar
Friday, April 15, 2011 - 14:00
1 hour (actually 50 minutes)
Skiles 005
University of California, San Diego, Department of Radiation Oncology
Cone Beam Computer tomography (CBCT) has been broadly applied incancer radiation therapy, mainly for positioning patients to align withtreatment radiation beams. As opposed to tomography reconstruction problemsfor diagnostic purposes, CBCT reconstruction in radiotherapy requires a highcomputational efficiency, since it is performed while patient is lying on acouch, waiting for the treatment. Moreover, the excessive radiation dosefrom frequent scans has become a clinical concern. It is therefore desirableto develop new techniques to reconstruct CBCT images from low dose scans. Inthis talk, I will present our recent work on an iterative low dose CBCTreconstruction technique via total variation regularization and tight frameregularization. It is found that 40~60 x-ray projections are sufficient toreconstruct a volumetric image with satisfactory quality in about 2min. Wehave also studied 4 dimensional CBCT (4DCBCT) reconstruction problem viatemporal non-local means (TNLM) and high quality 4DCBCT images can beobtained. Our algorithms have been fully implemented on a graphicsprocessing unit. Detailed implementation techniques will also be addressed.