Noise suppression-guided image filtering for low-SNR CT reconstruction

Area of Science:

• Medical Imaging
• Image Reconstruction
• Signal Processing

Background:

• Low signal-to-noise ratio (SNR) in computed tomography (CT) projections hinders image quality due to reduced radiation dose or device limits.
• Classical reconstruction algorithms like Simultaneous Algebraic Reconstruction Technique (SART) struggle with noisy CT data.
• Existing methods like POCS-BM3D can over-smooth images, losing crucial details during noise suppression.

Purpose of the Study:

• To develop an advanced CT reconstruction algorithm for low-SNR projections.
• To enhance the trade-off between noise suppression and edge preservation in CT image reconstruction.
• To improve the overall image quality of CT scans acquired with limited radiation dose or suboptimal hardware.

Main Methods:

• Introduced Guided Image Filtering (GIF) into the iterative CT reconstruction process.
• Developed the Noise Suppression-Guided Image Filtering Reconstruction (NSGIFR) algorithm.
• Integrated SART for initial reconstruction and BM3D for denoising, with GIF utilizing both for enhanced iterative refinement.

Main Results:

• The NSGIFR algorithm demonstrated superior performance in preserving structural details.
• NSGIFR effectively suppressed noise in low-SNR CT reconstructions.
• Quantitative and visual analyses confirmed NSGIFR's improved image quality over SART, POCS-TV, and POCS-BM3D.

Conclusions:

• NSGIFR offers a significant advancement in low-SNR CT reconstruction.
• The proposed method achieves a better balance between noise reduction and detail preservation.
• NSGIFR provides a valuable tool for improving CT imaging in dose-sensitive or hardware-constrained applications.