Dynamic PET Image Denoising Using Deep Image Prior Combined With Regularization by Denoising

Hao Sun,Hongyan Zhang,Yuru He,Shuangliang Cao,Lijun Lu,Lihong Peng

doi:10.1109/access.2021.3069236

Hao Sun, Hongyan Zhang + Show 4 more

Open Access

https://doi.org/10.1109/access.2021.3069236

Copy DOI

Abstract

The quantitative accuracy of positron emission tomography (PET) is affected by several factors, including the intrinsic resolution of the imaging system and inherently noisy data, which result in a low signal-to-noise ratio (SNR) of PET image. To address this problem, in this paper, we proposed a novel deep learning denoising framework aiming to enhance the quantitative accuracy of dynamic PET images via introduction of deep image prior (DIP) combined with Regularization by Denoising (RED), as such the method is labeled as DeepRED denoising. The network structure is based on encoder-decoder architecture and uses skip connections to combine hierarchical features to generate the estimated image. The network input can be random noise or other prior images (such as the patient’s own static PET image), avoiding the need of high quality noiseless images, which is limited in PET clinical practice due to high radiation dose. Based on simulated data and real patient data, the quantitative performance of the proposed method was compared with conventional Gaussian filtering (GF), non-local mean (NLM), block-matching and 3D filtering (BM3D), DIP and stochastic gradient Langevin dynamics (SGLD) method. Overall, the proposed method can outperform other conventional methods in substantial visual as well as quantitative accuracy improvements (in terms of noise versus bias performance) with and without prior images.

Highlights

Positron emission tomography (PET) is a nuclear medical imaging modality enabling quantitative measurements of biomolecular mechanisms by radioactive tracers in vivo
We develop a novel deep learning denoising framework aiming to enhance the quantitative accuracy of dynamic PET images via introduction of deep image prior combined with Regularization by Denoising (RED)
We evaluated the proposed DeepRED algorithm in comparison with conventional denoising methods, including Gaussian filtering (GF), non-local mean (NLM), block-matching and 3D filtering (BM3D), deep image prior (DIP) and stochastic gradient Langevin dynamics (SGLD) algorithms

Summary

Introduction

Positron emission tomography (PET) is a nuclear medical imaging modality enabling quantitative measurements of biomolecular mechanisms by radioactive tracers in vivo. Various pre and post-reconstruction algorithms have been developed by exploiting local statistics, spatiotemporal correlation, or prior anatomical information for PET image. The pre-processing method is carried out in PET sinogram to improve the estimation of physical parameters by using the spatiotemporal correlation in dynamic PET scans [5], [6]. For post-processing algorithms, conventional Gaussian filtering (GF) has been applied for PET image enhancement in clinic. Other post-processing algorithms, such as non-local mean (NLM) [7], wavelet [8], HYPR processing [9], guided image filtering [10], bilateral filtering [11] and kinetics-induced block matching and 5D filtering (KIBM5D) [12] have been proposed, and outperform conventional Gaussian filtering in terms of reducing image noise as well as preserving more image structure details

Methods

Results

Discussion

Conclusion