Nonlinear spatio-temporal filtering of dynamic PET data using a four-dimensional Gaussian filter and...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Signal Processing

Background:

• Dynamic Positron Emission Tomography (PET) imaging generates complex spatio-temporal data.
• Noise reduction is crucial for accurate quantitative analysis and interpretation of dynamic PET studies.
• Existing denoising methods have limitations in balancing noise suppression and signal preservation.

Purpose of the Study:

• To introduce and evaluate a novel denoising technique for dynamic PET data called Spatio-Temporal Expectation-Maximization (STEM) filtering.
• To assess the efficacy of STEM filtering in reducing variance in individual time frames and parametric images.
• To compare STEM filtering against established 3D and 4D denoising methods.

Main Methods:

• STEM filtering combines four-dimensional Gaussian filtering with Expectation-Maximization (EM) deconvolution.
• Gaussian filtering reduces noise across spatial and temporal frequencies.
• EM deconvolution rapidly restores signal-relevant frequencies.

Main Results:

• STEM filtering demonstrated substantial improvements in variance for individual PET time frames and parametric images.
• The method showed minimal bias, with early frame bias not impacting quantitative parameter estimates.
• STEM filtering outperformed other evaluated simple denoising techniques in dynamic phantom, simulated, and human PET studies.

Conclusions:

• STEM filtering is a simple yet effective method for denoising dynamic PET data.
• The technique offers significant advantages in reducing noise and improving quantitative accuracy in PET imaging.
• STEM filtering holds potential value for a broad spectrum of dynamic PET applications, including those using tracers with reversible and irreversible binding behaviors.