A General-Thresholding Solution for ℓp (0 < p <1) Regularized CT Reconstruction.

Area of Science:

• Signal Processing
• Image Reconstruction
• Applied Mathematics

Background:

• Compressive sensing utilizes ℓ1 minimization for sparse signal recovery.
• ℓp regularization (0 < p < 1) offers a generalized approach for sparser solutions.
• Existing methods like re-weighted algorithms have limitations in computed tomography (CT) reconstruction.

Purpose of the Study:

• To derive quasi-analytic thresholding representations for ℓp regularization.
• To develop and evaluate a general-threshold filtering algorithm for ℓp regularized CT reconstruction.
• To compare the proposed algorithm's performance against established methods.

Main Methods:

• Derivation of quasi-analytic thresholding representations for ℓp regularization (0 < p < 1).
• Analysis of error bounds for approximate general formulas.
• Integration of general-threshold formulas into an iterative thresholding framework for CT reconstruction.
• Evaluation using simulated and realistic data, including the Shepp-Logan phantom.

Main Results:

• The derived representations precisely match soft-thresholding (ℓ1) and hard-thresholding (ℓ0).
• The general-threshold filtering algorithm significantly reduces the required view number for accurate CT reconstruction.
• The proposed algorithm demonstrates superior performance in image quality, accuracy, convergence speed, and parameter sensitivity compared to the re-weighted approach.

Conclusions:

• The novel general-threshold filtering algorithm provides an effective and efficient solution for ℓp regularized CT reconstruction.
• This method advances sparse signal recovery techniques in medical imaging.
• The algorithm shows promise for improving CT reconstruction with reduced data acquisition.