Reduced iteration image reconstruction of incomplete projection CT using regularization strategy through Lp norm...

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Sparse and limited-angle Computed Tomography (CT) data often lead to significant artifacts in reconstructed images.
• Undersampled projection data is a primary cause of image degradation in CT.

Purpose of the Study:

• To develop and evaluate a novel reconstruction algorithm for sparse and limited-angle CT.
• To enhance image reconstruction quality by utilizing Dictionary Learning (DL) from sparse projections.

Main Methods:

• Proposed the ART-DL-Lp algorithm, integrating Dictionary Learning (DL) with the Algebraic Reconstruction Technique (ART).
• Employed signal sparse representation and feature extraction, constrained by L2 and Lp norms.
• Utilized an alternating strategy of 'ART first, then adaptive DL' for CT image reconstruction.
• Investigated the impact of different Lp norm values (0 < p < 1) on reconstruction performance.

Main Results:

• ART-DL-Lp demonstrated superior performance over ART, SART, and ART-DL-L2 algorithms with noisy, incomplete projections.
• Objective evaluation metrics (RMSE, MAE, PSNR, Residuals, SSIM) showed significant improvements with ART-DL-Lp.
• Lower p-values in ART-DL-Lp resulted in reconstructed images closer to the original and better objective metrics.

Conclusions:

• The ART-DL-Lp algorithm offers enhanced reconstruction efficiency for CT imaging with noisy, incomplete projections.
• ART-DL-Lp outperforms traditional ART, SART, and ART-DL-L2 methods in challenging CT scenarios.
• Optimizing the p-value in ART-DL-Lp is crucial for achieving the best reconstruction outcomes.