Super-Resolution Spiral Imaging Via Tensor Framelet: Megavoltage CT On TomoTherapy
Hao Gao* Departments of Mathematics and Computer Science, and Radiology and Imaging Sciences, Emory University, Atlanta, Georgia 30322 X. Sharon Qi Department of Radiation Oncology, University of California, Los Angeles, California, 90095SU-D-116-2 Sunday 2:05PM - 3:00PM Room: 116
Purpose: Spiral computed tomography (CT) acquires the sinogram projection views while the patient couch moves axially. As a result each projection corresponds to a different "narrow" axial slice. However, limited by the minimal requirement of ~180-degree projections, filtered backprojection (FBP) can only reconstruct an axially-averaged "thick" slice. This work is to investigate the recovery of multiple "narrow" axial slices per rotation for increasing the axial resolution, namely the super-resolution image reconstruction.
Methods: A key that enables the super-resolution image construction for spiral CT is the utility of the image slice similarity along the axial dimension, since the patient anatomy is often continuous with small variation within a narrow thickness. This prior of axial image coherence and the smoothness of each slice can be conveniently enforced by tensor framelet (TF), in which the image details are characterized by various filters at multilevel. In particular, we apply piecewise-linear TF that regularizes multilevel first-order and second-order gradients for x, y, and z directions in an isotropic fashion. The formulated convex L1-type optimization problem is efficiently solved by the split Bregman method with GPU accelerated X-ray transform and its adjoint.
Results: The validation of this super-resolution technique was performed using the setting of megavoltage (MV) CT on TomoTherapy with 1mm in-plane resolution and the couch speed 8mm per gantry rotation. The simulation results indicated that this TF-based super-resolution technique was able to reconstruct multiple slices per rotation, while FBP failed to do so. The experimental results suggested that this technique was able to reconstruct multiple slices with 1mm axial resolution, with new abundant information that was not available otherwise.
Conclusion: The super-resolution concept is introduced for spiral CT to enhance the axial image resolution. TF-based super-resolution reconstruction technique provides finer axial slices with 1mm axial resolution and in-plane resolution for TomoTherapy MVCT.
Funding Support, Disclosures, and Conflict of Interest: This work is partially supported by NIH/NIBIB grant EB013387.