Image restoration algorithm incorporating methods to remove noise and blurring from positron emission tomography imaging for Alzheimer's disease diagnosis

Abstract

The aim of this study was to design an image restoration algorithm that combined denoising and deblurring and to confirm its applicability in positron emission tomography (PET) images of patients with Alzheimer's disease (AD). PET images of patients with AD obtained using 18F-AV-45, which have a lot of noise, and 18F-FDG, which have a lot of blurring, were available in the Alzheimer's Disease Neuroimaging Initiative open dataset. The proposed framework performed image restoration incorporating blind deconvolution after noise reduction using a non-local means (NLM) approach to improve the PET image quality. We found that the coefficient of variation result after denoising and deblurring of the 18F-AV-45 image was improved 1.34 times compared to that for the degraded image. In addition, the profile result of the 18F-FDG PET image of patients with AD, which had a relatively large amount of blurring, showed a gentle shape when deblurring was performed after denoising. The overall no-reference-based evaluation results showed different results according to the degree of noise and blurring in the PET images. In conclusion, the applicability of the deconvolution deblurring algorithm to AD PET images after NLM denoising processing was demonstrated in this study.