Dynamic PET images denoising using spectral graph wavelet transform.

Area of Science:

• Medical Imaging
• Signal Processing
• Computational Biology

Background:

• Positron Emission Tomography (PET) provides quantitative molecular imaging but suffers from image quality degradation.
• Existing denoising methods like Gaussian filtering (GF) and non-local mean (NLM) often blur important edges, hindering lesion detection.
• Preserving edges during PET image denoising is critical for accurate diagnosis but remains a significant challenge.

Purpose of the Study:

• To develop a novel denoising method for dynamic PET images with superior edge-preserving capabilities.
• To address the limitations of conventional denoising techniques in maintaining high-frequency components essential for lesion recognition.

Main Methods:

• A composite image is generated from the dynamic PET time series.
• Spectral Graph Wavelet Transform (SGWT) is applied to the PET images.
• Denoising is achieved by reconstructing the low-frequency components of the graph representation.

Main Results:

• The proposed SGWT method significantly outperforms GF, NLM, and graph filtering in denoising dynamic PET images.
• The method demonstrates comparable denoising performance to deep learning approaches.
• The SGWT method exhibits lower computational complexity and does not require extensive training datasets.

Conclusions:

• The novel SGWT-based denoising method effectively enhances dynamic PET image quality while preserving critical edges.
• This approach offers a practical and efficient alternative to existing denoising techniques, including deep learning, for PET imaging analysis.