Denoising of 4D Cardiac Micro-CT Data Using Median-Centric Bilateral Filtration.

Area of Science:

• Medical Imaging
• Image Processing
• Computational Science

Background:

• Bilateral filtration is a key technique for denoising computed tomography (CT) data.
• Traditional methods use Gaussian weighting, but alternative distributions and higher dimensions offer potential improvements.
• Investigating novel weighting functions and filter variations is crucial for enhancing image quality.

Purpose of the Study:

• To evaluate various domain and range weighting distributions for bilateral filtration in CT denoising.
• To compare classic, median-centric, and hybrid bilateral filtration techniques.
• To determine the optimal filter for 4D cardiac micro-CT data denoising.

Main Methods:

• Brute-force optimization was used to test multiple weighting distributions (e.g., Flat, Lorentzian, Cosine).
• Median-centric and hybrid bilateral filters were developed to reduce bias in range weights.
• Filters were applied to a 4D MOBY mouse phantom and in vivo 4D cardiac micro-CT data.

Main Results:

• Hybrid bilateral filtration with Flat domain and range weighting yielded superior denoising on the MOBY phantom (PSNR 32.25).
• This hybrid filter also provided optimal smoothing in mouse left ventricle data (stdev 15.9 HU).
• The 3D hybrid filter is recommended for its computational tractability and artifact reduction in cardiac micro-CT.

Conclusions:

• Hybrid bilateral filtration with Flat weighting is highly effective for CT image denoising.
• The 3D hybrid filter balances performance and practicality for 4D cardiac micro-CT applications.
• This study advances CT denoising techniques through optimized bilateral filtration.