GPU-accelerated iterative reconstruction from Compton scattered data using a matched pair of conic projector and...

Area of Science:

• Medical Imaging
• Nuclear Physics
• Computer Science

Background:

• Compton scattered data reconstruction is computationally intensive due to conic projections.
• Conventional methods often use approximated, unmatched projector-backprojector pairs for GPU parallelization.
• Existing techniques face challenges in maintaining accuracy during iterative reconstruction.

Purpose of the Study:

• To develop a more accurate GPU-accelerated iterative reconstruction method for Compton cameras.
• To address the computational challenges of conic projections in Compton imaging.
• To utilize a precisely matched projector-backprojector pair for enhanced accuracy.

Main Methods:

• Employs a ray-tracing method (RTM) for conic forward projection, accumulating intersecting chord lengths.
• Utilizes a voxel-driven RTM for conic backprojection, ensuring it is the adjoint of the forward projector.
• Implements a matched pair of projectors for accurate iterative reconstruction on GPUs.

Main Results:

• The proposed method achieves reconstruction accuracy comparable to CPU-based methods.
• It is approximately 16 times faster than CPU-based reconstruction.
• The new method is about 3 times slower than approximated GPU methods but offers superior accuracy.

Conclusions:

• The perfectly matched projector-backprojector pair guarantees consistent reconstruction accuracy across iterations.
• This GPU-accelerated method significantly enhances the speed and accuracy of iterative reconstruction for Compton imaging.
• The findings pave the way for more efficient and reliable Compton camera applications.