PET Image Deblurring and Super-Resolution with an MR-Based Joint Entropy Prior

Area of Science:

• Medical Imaging
• Neuroimaging
• Image Processing

Background:

• Positron Emission Tomography (PET) suffers from limited spatial resolution, impacting quantitative accuracy.
• High-resolution anatomical information from Magnetic Resonance Imaging (MRI) can potentially improve PET image quality.

Purpose of the Study:

• To develop and validate an image deblurring and super-resolution framework for PET using anatomical guidance from MRI.
• To improve quantitative accuracy and image quality in PET scans.

Main Methods:

• A post-processing framework utilizing spatially-variant deconvolution of reconstructed PET images.
• Stabilization of deconvolution using an MRI-based joint entropy penalty function.
• Validation through simulations (BrainWeb phantom), experimental studies (Hoffman phantom), and clinical neuroimaging (Alzheimer's disease).

Main Results:

• Improved image quality and quantitative accuracy (CNR, SSIM, RMSE, PSNR) in BrainWeb simulations.
• Enhanced structural similarity index and gray-to-white contrast-to-noise ratio in Hoffman phantom studies.
• Reduced coefficient of variation in key brain regions for clinical Alzheimer's disease imaging.

Conclusions:

• The proposed framework effectively enhances PET image resolution and quantitative accuracy.
• This technique shows significant potential for improving neuroimaging studies, particularly in the context of aging and Alzheimer's disease.
• The MR-guided deconvolution method offers a valuable advancement over existing deconvolution techniques.