Deep denoiser prior driven relaxed iterated Tikhonov method for low-count PET image restoration

Area of Science:

• Medical Imaging
• Positron Emission Tomography (PET)
• Image Reconstruction

Background:

• Low-count PET imaging offers efficiency but suffers from noise and blurring.
• Existing PET image restoration methods have limitations like semi-convergence and lack of effective denoiser priors.

Purpose of the Study:

• To introduce a novel deep plug-and-play image restoration method for low-count PET.
• To address noise and blurring issues in low-count PET image reconstruction.

Main Methods:

• Developed Deep denoiser Prior driven Relaxed Iterated Tikhonov method (DP-RI-Tikhonov).
• Utilized a deep convolutional neural network denoiser for noise reduction.
• Implemented an iterative optimization algorithm with adaptive parameter selection for blurring reduction.

Main Results:

• DP-RI-Tikhonov achieved superior quantitative results (NRMSE, SSIM) compared to conventional and state-of-the-art methods.
• The method effectively reduced noise intensity and eliminated semi-convergence properties.
• Clinical experiments demonstrated sharper, more uniform PET images with recovered fine details.

Conclusions:

• DP-RI-Tikhonov overcomes limitations of existing PET image restoration techniques.
• The method shows significant potential for improving low-count PET imaging quality.
• This advancement could have broad implications for medical image restoration.