Wavelet denoising for voxel-based compartmental analysis of peripheral benzodiazepine receptors with 18F-FEDAA1106

Purpose: We evaluated the noise reduction capability of wavelet denoising for estimated binding potential (BP) images (k 3/k 4) of the peripheral benzodiazepine receptor using 18F-FEDAA1106 and nonlinear least-square fitting. Methods: Wavelet denoising within a three-dimensional discrete dual-tree complex wavelet transform was applied to simulate data and clinical dynamic positron emission tomography images of 18F-FEDAA1106. To eliminate noise components in wavelet coefficients, real and imaginary coefficients for each subband were thresholded individually using NormalShrink. A simulated dynamic brain image of 18F-FEDAA1106 was generated and Gaussian noise was added to mimic PET dynamic scan. The derived BP images were compared with true images using 156 rectangular regions of interest. Wavelet denoising was also applied to data derived from seven young normal volunteers. Results: In the simulations, estimated BP by denoised image showed better correlation with the true BP values (Y=0.83X+0.94, r=0.80), although no correlation was observed in the estimates between noise-added and true images (Y=1.2X+0.78, r=0.49). For clinical data, there were visual improvements in the signal-to-noise ratio for estimated BP images. Conclusions: Wavelet denoising improved the bias and reduced the variation of pharmacokinetic parameters for BP