A non-local means post-filter with spatially adaptive filtering strength for whole-body PET

We propose a spatially adaptive Non-Local Means (NLM) post-filtering approach for whole-body clinical Positron Emission Tomography (PET) imaging. Our approach is aimed at avoiding different effective smoothing strengths in different organs that result from with traditional non-adaptive NLM. We vary the smoothing strength according to the intensity level around a given voxel such that regions with low absolute noise levels are smoothed less and those with high absolute noise levels are smoothed more. We evaluated this approach and compared it to alternative filtering techniques by inserting lesions of known size and activity into clinical datasets acquired on a Toshiba Celesteion
^TM
TOF-PET/CT scanner. Images were reconstructed with list-mode Ordered-Subset Expectation-Maximization (OSEM) algorithm with all physical corrections inside the system model prior to post-filtering. We qualitatively evaluated the techniques by comparing the differences between filtered and original images and quantitatively evaluated them using lesion contrast vs. background variability curves. Both evaluations showed that the proposed method could better accommodate varying noise statistics in whole-body PET images and better preserve lesion contrast across different anatomical regions.