Few-view CT reconstruction with group-sparsity regularization.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Classical total variation-based algorithms cause oversmoothing in textured regions of reconstructed images.
• Few-view computed tomography (CT) reconstruction is challenging due to limited data.

Purpose of the Study:

• To develop a novel CT reconstruction method to address the oversmoothing issue in textured regions for few-view problems.
• To improve the quantitative and qualitative performance of CT image reconstruction.

Main Methods:

• A group-sparsity regularization-based simultaneous algebraic reconstruction technique (SART) was developed.
• Group-based sparse representation using nonlocal patches clustered by Euclidean distance was employed as a prior regularization term.
• The split Bregman iteration algorithm was used for numerical implementation.

Main Results:

• The proposed method effectively eliminates the oversmoothing effect in textured regions.
• Qualitative and quantitative experiments showed superior performance compared to filtered back projection, SART, total variation-based projections onto convex sets, and SART-based dictionary learning.
• The method simultaneously explores sparsity and nonlocal similarity within a single image.

Conclusions:

• The group-sparsity regularization-based SART method offers a significant improvement for few-view CT reconstruction, particularly for images with textured regions.
• This approach provides a more accurate and detailed reconstruction than traditional methods.
• The findings suggest a promising direction for enhancing CT imaging quality in data-limited scenarios.