A novel approach for CT image smoothing: Quaternion Bilateral Filtering for kernel conversion

Area of Science:

• Medical Imaging
• Image Processing
• Computational Mathematics

Background:

• Denoising reconstructed Computed Tomography (CT) images without raw projection data is challenging, especially with sharp kernels causing high-frequency noise.
• Existing methods often require raw data or rely on grayscale filtering, limiting their effectiveness on already reconstructed images.

Purpose of the Study:

• To develop an innovative method for denoising sharp kernel CT images post-reconstruction.
• To improve the quality of CT images by mitigating kernel-induced noise while preserving anatomical structures.

Main Methods:

• Developed a Quaternion Bilateral Filter (QBF) integrating quaternion mathematics with bilateral filtering.
• Expressed CT scans in quaternion form, encoding RGB channels together for unified processing.
• Applied the QBF directly to reconstructed sharp kernel images, using paired data to mimic soft-kernel outputs.

Main Results:

• The QBF demonstrated superior denoising performance compared to traditional Bilateral Filter (BF), Non-Local Means (NLM), wavelet, and Convolutional Neural Network (CNN) methods.
• Achieved high performance metrics, including a Structural Similarity Index Measure (SSIM) of 0.96 and Peak Signal-to-Noise Ratio (PSNR) of 36.3 on B50f reconstructions.
• Segmentation-based validation confirmed that QBF-filtered images retain essential structural details for diagnostics.

Conclusions:

• The Quaternion Bilateral Filter (QBF) is an effective and efficient method for post-reconstruction CT image denoising.
• QBF offers a lightweight, interpretable alternative to deep learning models for enhancing CT image quality.
• This approach successfully addresses the challenge of denoising without raw projection data, improving diagnostic utility.