Convolutional Sparse Coding for Compressed Sensing CT Reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Dictionary learning (DL) methods have advanced image reconstruction but suffer from patch-based limitations, ignoring pixel consistency and learning redundant features.
• Traditional DL methods in computed tomography (CT) can introduce artifacts due to patch aggregation and fail to preserve fine details.

Purpose of the Study:

• To explore the application of convolutional sparse coding (CSC) for sparse-view CT reconstruction.
• To address the limitations of patch-based DL methods in CT image reconstruction.

Main Methods:

• Developed a CSC-based approach for sparse-view CT reconstruction that operates on the entire image.
• Utilized predetermined filters and an alternating optimization scheme to refine the objective function.
• Validated the method using both simulated and real CT data.

Main Results:

• The proposed CSC method preserves more image details and avoids artifacts common in patch-based DL methods.
• Experiments demonstrated superior performance compared to existing state-of-the-art methods in qualitative and quantitative evaluations.
• The CSC approach effectively handles sparse-view CT reconstruction challenges.

Conclusions:

• Convolutional sparse coding offers a significant improvement over traditional dictionary learning for sparse-view CT reconstruction.
• The whole-image processing nature of CSC enhances image quality and reduces artifacts.
• The developed CSC method presents a promising alternative for high-quality CT image reconstruction.