Single-scan adaptive graph filtering for dynamic PET denoising by exploring intrinsic spatio-temporal structure.

Area of Science:

• Medical Imaging
• Signal Processing
• Computational Science

Background:

• Dynamic Positron Emission Tomography (PET) performance suffers from high noise levels and scan variability.
• Existing denoising methods often use fixed models or are limited to single data domains, hindering structural information capture.

Purpose of the Study:

• To introduce a novel single-scan adaptive spatio-temporal graph filtering (ST-GF) technique for PET image denoising.
• To overcome limitations of current denoising algorithms by developing a method that adapts to individual scan characteristics.

Main Methods:

• Representing noisy PET sinogram data in a graph-signal space to explore latent spatio-temporal structures.
• Developing an iterative framework with an adaptively constructed graph filter tailored to each scan's unique characteristics.
• Applying the ST-GF technique directly to single acquisitions without requiring external training data.

Main Results:

• The ST-GF technique effectively reveals underlying spatio-temporal structures obscured by noise.
• Experiments on simulated and in vivo datasets demonstrated superior denoising performance compared to existing methods.
• The method successfully separates true signal from noise by adapting to the specific characteristics of each scan.

Conclusions:

• The ST-GF technique provides robust and flexible denoising for Dynamic PET by leveraging scan-specific latent structures.
• Operating without prior training, the method is immune to external data biases, showing significant potential for practical applications.
• This adaptive approach enhances image quality and reliability in Dynamic PET imaging.