GPU-accelerated regularized iterative reconstruction for few-view cone beam CT.

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiology

Background:

• X-ray transmission computed tomography (CT) is crucial for medical diagnosis.
• Low-dose and few-view acquisitions are desirable to minimize patient radiation exposure and scan time.
• Reconstructing high-quality images from limited data remains a significant challenge in CT.

Purpose of the Study:

• To propose a model-based iterative reconstruction technique for cone-beam CT (CBCT).
• To achieve accurate low-dose image reconstruction from few-view acquisitions.
• To ensure reconstruction is performed within clinically acceptable time frames.

Main Methods:

• The study developed an Ordered Subsets Convex (OSC) algorithm combined with Total Variation (TV) minimization, termed OSC-TV.
• A reduction pattern for the number of subsets in OSC iterations was implemented to optimize regularization.
• The computationally intensive algorithm was parallelized and implemented on a graphics processing unit (GPU) for acceleration.

Main Results:

• OSC-TV demonstrated effective reconstruction of low-dose, few-view CBCT data, showing good convergence and image quality.
• Performance was evaluated using simulated and human pelvic CBCT data.
• OSC-TV compared favorably to the POCS-TV algorithm and offered a viable alternative to filtered backprojection, with execution times of 1-2 minutes.

Conclusions:

• The OSC-TV method provides high-quality reconstructions suitable for low-dose clinical CT acquisitions.
• Its efficiency and accuracy make it beneficial for patients requiring multiple imaging procedures, reducing radiation risks.
• This technique represents a significant advancement in low-dose CT imaging, balancing image fidelity with patient safety.