Combining ordered subsets and momentum for accelerated X-ray CT image reconstruction.

Area of Science:

• Medical Imaging
• Computational Science
• Image Reconstruction

Background:

• Statistical X-ray computed tomography (CT) reconstruction enhances image quality from low-dose scans.
• Ordered subsets (OS) methods are common in CT, PET, and SPECT reconstruction.
• OS methods based on separable quadratic surrogates (OS-SQS) are parallelizable but require many iterations for convergence.

Purpose of the Study:

• To introduce novel algorithms that accelerate statistical X-ray CT reconstruction.
• To improve the convergence speed of OS-SQS methods for practical applications.
• To enhance the stability of accelerated reconstruction methods.

Main Methods:

• Developed OS-SQS-momentum algorithms combining Nesterov's momentum with OS-SQS.
• Implemented diminishing step sizes to stabilize the algorithms when using a large number of subsets.
• Validated performance using simulated and real 3D CT scan data.

Main Results:

• The proposed OS-SQS-momentum algorithms significantly improve convergence speed in early iterations.
• Diminishing step sizes stabilize the method without compromising fast convergence.
• Experimental results demonstrate the effectiveness of the new algorithms on CT data.

Conclusions:

• OS-SQS-momentum algorithms with diminishing step sizes provide a substantial speedup for statistical CT reconstruction.
• These methods offer a practical solution for achieving high-quality, low-dose CT images more efficiently.
• The approach holds promise for improving clinical workflows in X-ray CT imaging.