A Non-Local Means Approach for PET Image Denoising

Abstract

Objective To explore a new method for removing the noise in PET images efficiently with the important diagnosis structures preserved. Method The Non-local means algorithm appeared recently was adapted for PET image denoising, one synthetic test image with poisson noise added and two real clinical PET images were used to validate the denoising results. Then the results were compared with those of traditional median filter and wiener filter. Results Experimental results for a test image and two real PET images show that Non-local means method is superior to median filtering and wiener filtering methods. For the test image with poisson noise added, the peak signal to noise ratio was increased by 10.9dB; For clinical PET images, the obtained images have more clearer boundaries and more visible diagnosis structures. Conclusions The Non-local means algorithm can suppress noise in PET images effectively with important structure details for diagnosis preserved and provide a new approach for denoising PET images.