3D alternating direction TV-based cone-beam CT reconstruction with efficient GPU implementation.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Iterative image reconstruction (IIR) methods, like total variation (TV) minimization, offer reduced sampling but face high computational complexity, particularly for 3D cone-beam computed tomography (CBCT).
• Existing IIR algorithms struggle with the computational burden of large matrix operations in 3D CBCT, limiting their practical application.

Purpose of the Study:

• To develop an efficient and accurate iterative image reconstruction algorithm for 3D cone-beam computed tomography (CBCT).
• To leverage GPU implementation for substantial acceleration of the proposed CBCT reconstruction algorithm.

Main Methods:

• Proposed an algorithm based on alternating direction total variation (ADTV) incorporating local linearization and a proximity technique for CBCT reconstruction.
• The proximity technique effectively bypasses computationally intensive pseudoinverse calculations of large matrices.
• Implemented the algorithm on a Graphics Processing Unit (GPU) for enhanced computational performance.

Main Results:

• The proposed ADTV algorithm demonstrates a simple yet convergent iterative process for CBCT.
• Achieved significant speed improvements with a GPU implementation, showing an acceleration ratio exceeding 100x compared to CPU computation.
• Reconstruction of 3D images (256x256x256) with 36 projections (512x512) took approximately 6.8 seconds per iteration, maintaining numerical accuracy.

Conclusions:

• The developed GPU-accelerated ADTV algorithm provides an efficient and accurate solution for 3D CBCT reconstruction.
• This approach overcomes the computational bottlenecks of traditional IIR methods, making it suitable for practical CBCT imaging applications.
• The substantial acceleration achieved by GPU implementation enables faster and more effective medical imaging.