Statistical Interior Tomography via L₁ Norm Dictionary Learning without Assuming an Object Support.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Interior tomography in X-ray computed tomography (CT) offers reduced radiation dose and hardware costs.
• Truncated projection data in interior CT limits traditional reconstruction algorithms.
• Existing methods struggle with artifacts and noise, hindering clinical application.

Purpose of the Study:

• To develop a high-quality statistical iterative reconstruction algorithm for interior CT.
• To address limitations caused by truncated projection data.
• To improve image quality by reducing artifacts and noise.

Main Methods:

• Proposed a statistical iterative reconstruction algorithm incorporating zeroth-order image moment prior knowledge.
• Estimated zeroth-order image moment in the projection domain using the Helgason-Ludwig consistency condition.
• Incorporated L1-norm sparse representation (dictionary learning) and moment constraints into the objective function.
• Minimized the objective function using an alternating minimization iterative algorithm.

Main Results:

• The proposed algorithm effectively removes shift artifacts.
• Demonstrated superior performance in noise reduction compared to total variation (TV)-based methods.
• Showcased enhanced preservation of fine structures in simulated and real CT data.

Conclusions:

• The novel algorithm provides high-quality interior CT reconstruction.
• It overcomes limitations of traditional methods dealing with truncated data.
• Offers improved diagnostic potential for clinical applications through enhanced image quality.