Mutual Information-Based Non-Local Total Variation Denoiser for Low-Dose Cone-Beam Computed Tomography

Area of Science:

• Medical Imaging
• Image Processing
• Radiotherapy

Background:

• Conventional non-local total variation (NLTV) methods struggle with low-dose cone-beam computed tomography (CBCT) due to noisy patches.
• Non-local means (NLM) filter weights degrade in low milliampere-seconds (mAs) CBCT images, impacting similarity calculations.
• Existing methods often fail to preserve fine details in low-dose CBCT reconstructions.

Purpose of the Study:

• To introduce a novel mutual information-based non-local total variation (MI-NLTV) denoising method for low-dose CBCT.
• To enhance the stability and robustness of NLTV in low-dose CBCT imaging.
• To improve the clinical image quality of CBCT acquired with low mAs.

Main Methods:

• Developed MI-NLTV using mutual information (MI) for similarity calculation between non-local and reference patches.
• Applied MI-NLTV denoising to CBCT images reconstructed with an analytical algorithm and ray-driven backprojector (RDB).
• Minimized the MI-NLTV objective function using steepest gradient descent optimization.

Main Results:

• MI-NLTV demonstrated superior performance compared to conventional NLTV in low-dose CBCT.
• Achieved significant improvements: 15.97% higher contrast-to-noise ratio, 2.67% lower RMSE, 0.12% lower non-uniformity, 1.14% higher correlation, and 18.11% higher detectability index.
• The integrated RDB further enhanced the MI-NLTV denoising algorithm's effectiveness.

Conclusions:

• MI-NLTV offers a more stable and robust solution for low-dose CBCT denoising than conventional NLM filters.
• The method preserves fine structures and details in reconstructed images.
• Achieving clinically acceptable CBCT quality at low mAs reduces patient burden and improves safety in radiotherapy.