[Paper Review] GPU-based Fast Low-dose Cone Beam CT Reconstruction via Total Variation

Xun Jia, Yifei Lou+6 more|arXiv (Cornell University)|Jan 4, 2010

Medical Imaging Techniques and Applications29 references4 citations

TL;DR

This paper presents a GPU-accelerated iterative reconstruction algorithm for low-dose cone-beam CT (CBCT) using total variation (TV) regularization to reduce radiation exposure. By leveraging a GPU-friendly forward-backward splitting algorithm with multi-grid acceleration, the method achieves high-quality CBCT reconstruction from as few as 40 projections at 0.1 mAs/projection, reducing dose by up to 36 times while reconstructing in ~130 seconds—100× faster than conventional iterative methods.

ABSTRACT

Cone-beam CT (CBCT) has been widely used in image guided radiation therapy (IGRT) to acquire updated volumetric anatomical information before treatment fractions for accurate patient alignment purpose. However, the excessive x-ray imaging dose from serial CBCT scans raises a clinical concern in most IGRT procedures. The excessive imaging dose can be effectively reduced by reducing the number of x-ray projections and/or lowering mAs levels in a CBCT scan. The goal of this work is to develop a fast GPU-based algorithm to reconstruct high quality CBCT images from undersampled and noisy projection data so as to lower the imaging dose. The CBCT is reconstructed by minimizing an energy functional consisting of a data fidelity term and a total variation regularization term. We developed a GPU-friendly version of the forward-backward splitting algorithm to solve this model. A multi-grid technique is also employed. We test our CBCT reconstruction algorithm on a digital NCAT phantom and a head-and-neck patient case. The performance under low mAs is also validated using a physical Catphan phantom and a head-and-neck Rando phantom. It is found that 40 x-ray projections are sufficient to reconstruct CBCT images with satisfactory quality for IGRT patient alignment purpose. Phantom experiments indicated that CBCT images can be successfully reconstructed with our algorithm under as low as 0.1 mAs/projection level. Comparing with currently widely used full-fan head-and-neck scanning protocol of about 360 projections with 0.4 mAs/projection, it is estimated that an overall 36 times dose reduction has been achieved with our algorithm. Moreover, the reconstruction time is about 130 sec on an NVIDIA Tesla C1060 GPU card, which is estimated ~100 times faster than similar iterative reconstruction approaches.

Motivation & Objective

• To address the clinical challenge of excessive radiation dose from repeated CBCT scans in image-guided radiation therapy (IGRT).
• To enable high-quality CBCT reconstruction from undersampled and noisy projection data acquired at low dose.
• To develop a fast, GPU-accelerated algorithm that maintains image quality while drastically reducing imaging dose.

Proposed method

• Formulates CBCT reconstruction as minimizing an energy functional combining data fidelity and total variation (TV) regularization.
• Develops a GPU-optimized forward-backward splitting algorithm to efficiently solve the TV-regularized optimization problem.
• Employs a multi-grid technique to accelerate convergence and improve reconstruction speed on GPU architecture.
• Uses a digital NCAT phantom, patient anatomy, and physical Catphan and Rando phantoms for validation.
• Implements iterative reconstruction with TV minimization to suppress noise and preserve edges in low-dose data.
• Optimizes computational performance by exploiting GPU parallelism and memory coalescing.

Experimental results

Research questions

• RQ1 Can total variation regularization effectively reconstruct high-quality CBCT images from severely undersampled and noisy projection data at low dose?
• RQ2 How much dose reduction is achievable while maintaining diagnostic image quality for patient setup in IGRT?
• RQ3 Can GPU acceleration reduce reconstruction time to clinically feasible levels compared to conventional iterative methods?
• RQ4 What is the minimum number of projections and mAs level required for diagnostically useful CBCT reconstruction?
• RQ5 How does the multi-grid technique enhance reconstruction speed and convergence in GPU-based iterative reconstruction?

Key findings

• As few as 40 x-ray projections at 0.1 mAs/projection produced CBCT images of sufficient quality for patient alignment in IGRT.
• The method achieved an estimated 36-fold dose reduction compared to the standard full-fan head-and-neck protocol (360 projections at 0.4 mAs/projection).
• Reconstruction time was approximately 130 seconds on an NVIDIA Tesla C1060 GPU, representing a ~100× speedup over conventional iterative approaches.
• Physical phantom experiments confirmed successful image reconstruction at the ultra-low dose level of 0.1 mAs/projection.
• The GPU-optimized algorithm maintained image fidelity and edge preservation while effectively suppressing noise in low-dose data.
• The multi-grid technique significantly improved convergence speed without compromising image quality.