Graph Filtering Approach to PET Image Denoising

Positron emission tomography (PET) is an important biomedical imaging modality for disease, eg cancer, diagnosis. Its accuracy is often limited by high level of noise relative to signal due to constraints on injected dose and signal acquisition time. To denoise PET images, various denoising methods, such as nonlocal means (NLM) and block-matching and 3D (BM3D) filtering, have been used. Because PET images have high contrast and low spatial resolution, these denoising methods do not reduce the noise effectively while preserving the image edges. This paper proposes a novel approach to PET image denoising based on graph filtering. Since the graph spectrum distributions of image signals and noise in PET image are different, the proposed approach reduces the noise in PET image by graph spectral filtering. Simulated, preclinical and clinical datasets are used to evaluate the performance of the proposed approach. Compared with NLM and BM3D filtering, the proposed approach improves the peak signal-to-noise ratio and structural similarity index for the simulated dataset, and subjectively reduces the noise while preserving the image edges and fine details for the preclinical and clinical datasets.

I. Introduction

PET is a powerful functional imaging modality. It provides diagnostic by measuring tracer activity. It has been used for various diseases, including neurological disorders, infection, and inflammation. In radiation oncology applications, for instance, PET helps localization and staging of the tumors. PET images have high contrast, but low spatial resolution compared to magnetic resonance imaging (MRI) and computed tomography (CT). Since PET image is constructed via sinogram, the reconstructed PET image has inevitable noise. The noisy data greatly affect the accuracy of doctor diagnosis. Hence, PET image denoising is necessary.