Wavelet Denoising of Dynamic PET Data: Application to the Parametric Imaging of Peripheral Benzodiazepine Receptor

Introduction: Low signal-to-noise ratio (SNR) of time activity curve (TAC) at the voxel level causes severe bias and poor precision for estimated binding potential (BP) in peripheral benzodiazepine receptor (PBR) using Nonlinear least square (NLS) fitting. The purpose of this study is to evaluate noise reduction capability of wavelet denoising for estimated BP image. Methods: We applied spatial (3D) wavelet denoising to simulate data and clinical dynamic image of PBR using 18F-FEDAA1106. For denoising process, wavelet coefficient for each sub-band was thresholded by soft-threshold method. In simulation study, we generated human mimicked dynamic phantom images and then evaluated the bias of BP from denoised image compared with that from true BP image. In clinical application, BP image with/without wavelet denoising for a subject were compared. Results: Overestimation of BP values was 10~13% in case of noise level 10% of simulation. Wavelet denoising improved overestimated BP values almost a few percent. In clinical, wavelet denoising can compensate the noise in dynamic image, however, difference between BP images of with/without wavelet denoising were not so apparent. Conclusion: Wavelet denoising improve the quality of dynamic images, however, further optimization will be required for improving SNR of BP images.