Noise adaptive multi-resolution technique to accurately denoise PET, MRI-PET, and PET-CT images

Objectives 1. To develop a fast and robust method for automatically denoising PET, MRI-PET, and PET-CT images based on an accurate statistical noise models in the multi-resolution domain. 2. To preserve crucial information such as SUV based quantitative metrics, and object boundary information while denoising.

Methods With IRB approval, 20 co-registered PET-CT, 20 MRI-PET, and 27 PET images were retrospectively collected. Our proposed denoising method consists of three steps: (1) a novel adaptive multi-scale approach to denoise three unique imaging modalities that are corrupted by three different noise models (PET corrupted by Poisson-Gaussian noise, MR images corrupted by Rician-distributed noise, and CT images corrupted by white Gaussian noise), (2) edge-preserving counterlet transform was used to denoise transformed images, and (3) an optimal inverse multi-scale transform to get the denoised images in SUV domain.

Results Significant improvement in signal to noise ratio (SNR) (18.57 vs. 16.68, p<0.01) and relative contrast ratio (see Fig. 1B (f)) were obtained. Furthermore, clinically significant values such as SUVmax and SUVmean values changed after the denoising were minimized within the scope of the clinical limit (i.e., max 10% change was observed). Figure 1B shows sample PET images, before and after noise was removed. We observed that noise stabilization step is often ignored in the literature and leading to inaccurate results with losing crucial clinical information. Further we extended our approach to denoise MR images for which noise was considered to be Rician distributed. To stabilize the variance in MR images; we define a unique MS-VST function adapted to the Rician noise model. For CT images, conventional Gaussian noise assumption was used for denoising along with the proposed method without using a stabilization step beforehand..

Conclusions The preliminary study showed the feasibility of a fast, accurate, and automated method for denoising, PET, MRI-PET, and PET-CT images.

Research Support This research is supported by Center for Infectious Disease Imaging (CIDI), the intramural research program of the National Institutes of Allergy and Infectious Diseases (NIAID) and the National Institute of Biomedical Imaging and Bioengineering (NIBIB).