Kernel graph filtering-A new method for dynamic sinogram denoising.

Area of Science:

• Medical Imaging
• Signal Processing
• Computational Science

Background:

• Low count positron emission tomography (PET) imaging is crucial for clinical diagnosis and research.
• High noise levels in low count PET images, stemming from weak sinogram signals, hinder accurate image reconstruction.
• Existing denoising methods often struggle to effectively enhance image quality, particularly at low signal levels.

Purpose of the Study:

• To develop and evaluate a novel kernel graph filtering method for dynamic PET sinogram denoising.
• To improve the signal-to-noise ratio and overall image quality in low count dynamic PET imaging.
• To provide a robust solution for enhancing dynamic PET data acquisition and analysis.

Main Methods:

• A novel kernel graph filtering approach was developed, treating dynamic sinograms as graph signals.
• The graph structure was adaptively learned using kernel principal components of the sinograms.
• A lowpass kernel graph spectrum filter was constructed and applied to denoise sinogram time frames.

Main Results:

• The proposed kernel graph filtering method demonstrated superior performance in sinogram denoising compared to existing techniques.
• Significant image enhancement was observed in dynamic PET datasets across all count levels, particularly at low counts.
• The method effectively reduced noise while preserving important signal information for PET image reconstruction.

Conclusions:

• The novel kernel graph filtering method offers a significant advancement in dynamic PET sinogram denoising.
• This technique holds great potential for improving the clinical utility and research applications of low count dynamic PET imaging.
• The method provides a valuable tool for enhancing image quality and diagnostic accuracy in challenging low-count PET scenarios.