A novel partial volume effects correction technique integrating deconvolution associated with denoising within an...

Area of Science:

• Medical Imaging
• Nuclear Medicine
• Image Reconstruction

Background:

• Partial volume effect (PVE) significantly impacts PET image accuracy, particularly for small structures.
• Existing voxelwise PVE correction methods use post-reconstruction Lucy-Richardson deconvolution with wavelet denoising.
• Noise propagation remains a challenge in current PVE correction techniques.

Purpose of the Study:

• To integrate Lucy-Richardson deconvolution and wavelet-based denoising directly into the iterative PET image reconstruction process.
• To improve partial volume effect (PVE) correction accuracy and efficiency.
• To evaluate the performance of in-reconstruction PVE correction compared to post-reconstruction methods.

Main Methods:

• Modified the list-mode ordered subset expectation maximization (OSEM) algorithm to include Lucy-Richardson deconvolution at each iteration.
• Evaluated the method on NEMA IQ phantom data acquired on a GE DRX PET/CT system.
• Assessed image quality by analyzing intensity recovery versus background noise (relative standard deviation) and performing qualitative clinical case reviews.

Main Results:

• In-reconstruction deconvolution without denoising yielded superior intensity recovery compared to standard OSEM with PSF and post-reconstruction deconvolution.
• Incorporating wavelet-based denoising effectively limited signal-to-noise ratio (SNR) degradation while maintaining intensity recovery.
• The integrated approach demonstrated improved PVE correction within the iterative reconstruction.

Conclusions:

• Integrating Lucy-Richardson deconvolution with wavelet denoising into the PET reconstruction process is feasible and effective for PVE correction.
• The proposed method offers enhanced accuracy for quantitative PET imaging, especially in small structures.
• Future research will focus on clinical dataset evaluation and advanced point spread function (PSF) models.