SU-F-BRCD-09: Total Variation (TV) Based Fast Convergent Iterative CBCT Reconstruction with GPU Acceleration

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiation Oncology

Background:

• Cone Beam CT (CBCT) is crucial for radiation therapy, but image quality and radiation dose are significant concerns.
• Current reconstruction methods like FDK may not fully address artifacts from limited projection data.

Purpose of the Study:

• To develop and validate a fast, iterative image reconstruction algorithm for CBCT.
• To enhance image quality and reduce radiation dose in radiation therapy applications.
• To achieve near real-time reconstruction using multi-GPU acceleration.

Main Methods:

• An iterative algorithm minimizing a weighted least squares cost function with total variation (TV) regularization was implemented.
• A multi-GPU implementation was optimized for rapid 3D reconstruction (< 1 minute).
• Algorithm performance was assessed using digital (Shepp-Logan) and physical (Catphan-600) phantoms, comparing results to FDK.

Main Results:

• The iterative algorithm achieved convergence in as few as 15 iterations for 360-view and 60 iterations for 60-view cases.
• Root Mean Square Error (RMSE) was significantly reduced.
• Multi-GPU acceleration enabled reconstruction in under 5.4 seconds per iteration.
• Iterative reconstructions showed superior spatial resolution and Signal-to-Noise Ratio (SNR) compared to FDK.

Conclusions:

• A fast-converging iterative algorithm for CBCT reconstruction was successfully developed.
• The algorithm produces higher quality images (spatial resolution, SNR) than commercial FDK methods.
• The approach shows potential for significant radiation dose reduction in few-view CBCT scenarios.
• This method is expected to benefit image-guided adaptive radiation therapy (IGART) and daily patient localization.