A Limited-View CT Reconstruction Framework Based on Hybrid Domains and Spatial Correlation.

Area of Science:

• Medical Imaging
• Image Reconstruction
• Computational Imaging

Background:

• Limited-view Computed Tomography (CT) reduces radiation dose but introduces severe artifacts.
• Existing methods often neglect spatial correlations between consecutive CT images.
• High-quality image reconstruction from limited data is crucial for diagnosis and inspection.

Purpose of the Study:

• To develop a novel method for high-quality CT image reconstruction from limited-view data.
• To exploit spatial correlations between continuous CT images for artifact reduction.
• To improve the efficiency of limited-view CT reconstruction.

Main Methods:

• A hybrid-domain structure using fully convolutional networks.
• Exploration of three-dimensional neighborhood in a "coarse-to-fine" manner.
• Radon domain data completion followed by Filtered Back Projection (FBP) image reconstruction.
• Utilizing spatial correlations for image restoration and texture refinement.
• GPU-accelerated FBP for efficient end-to-end reconstruction.

Main Results:

• Achieved high-quality CT images with PSNR of 40.209 and SSIM of 0.943.
• Significantly reduced artifacts inherent in limited-view CT.
• Demonstrated effective exploitation of spatial information between CT images.
• Accelerated the overall reconstruction procedure compared to traditional methods.

Conclusions:

• The proposed hybrid-domain network effectively reconstructs high-quality CT images from limited-view data.
• Leveraging spatial correlations is a promising approach for limited-view CT artifact reduction.
• The GPU-accelerated FBP method offers an efficient and end-to-end solution for limited-view CT reconstruction.