The De-noising Methods for SPECT Reconstruction with Non-uniform Attenuation Based on Wavelet Transform

In single photon emission computed tomography (SPECT), the non-stationary Possion noise in the projection data (sinogram) is a major cause to compromise the quality of the reconstructed images. To improve the quality, we must remove the Possion noise in the sinogram before reconstruction. However, the conventional space or frequency domain de-noising methods possibly remove the edge information that is very important for the accurate reconstruction, especially for SPECT reconstruction with non-uniform attenuation. As a time-frequency analysis tool, wavelet transform has been widely used in the signal and image processing fields, demonstrated its powerful functions in the application of de-noising. In this paper, we try to find out the de-noising abilities of the wavelet based de-noising methods for SPECT reconstruction with non-uniform attenuation, and the effect of the Anscombe transform in the wavelet based de-noising. Five most effective de-noising methods were selected, and the non-uniform attenuation reconstruction algorithm was applied to the de-noised projection data. From the reconstructed results, it is clear that the Revised BivaShrink with complex wavelet is the best wavelet based de-noising method for SPECT reconstruction with non-uniform attenuation, and the effect of Anscombe transform, which converts the Possion noise into Gaussian noise, is not significant in the wavelet based de-noising.