MR-guided PET image denoising

With the growing interest in conducting multi-centre and multi-modality studies on neurological disorders, post-reconstruction PET image enhancement methods that take advantage of available anatomical information are becoming more important. In this work, a novel method for denoising PET images using the subject's registered T1-weighted MR image is proposed. The proposed method combines the non-local means approach with the twicing strategy from the image denoising literature to restore a reconstructed PET image. Preliminary analysis shows promising improvements in peak signal to noise ratio (PSNR) and contrast recovery coefficients (CRC) of the lesions when denoising simulated images reconstructed using the MLEM algorithm.