PET denoising and deconvolution using prior information from CT or MRI

PET examination is a well established method in medicine, mostly for identifying lesions as being malignant or benign. Surgery for the lesion is almost always based on structural examination provided by MRI or CT. A problem for surgery appears if the existing pathological tissue is not identifiable by MRI/CT but can be seen by PET (as for example in some cases of epilepsy patient). As the PET scans are structurally not precise enough it is difficult for neurosurgeon to identify the brain lesion and there is a need for an invasive exploration of the brain, hence increasing the perioperative morbidity. Making the PET image more focused and less noisy would decrease the risk of surgery.


We have proposed a denoising and blind deconvolution MAP method for PET images, which includes structural information from MRI/CT images as a modified total variation prior distribution. We call the modified prior structural total variation (STV). The proposed method provides better focused PET images and it is robust to misalignment.


An example of one slice of the phantom volume acquired in both modalities (PET and CT) is given below in Fig. 1. The phantom contains cold rods of different dimensions positioned into triangles oriented around the center.


Fig. 1. Example of a phantom volume acquired by a combined PET/CT scanner: (left) CT (512x512x159) image and (right) corresponding PET (256x256x47) image.


The next Fig. 2 shows reconstruction by the state-of-the-art wavelet-based method (see reference in Relevant Literature) and by our STV method. Both methods give equal results in terms of denoising. However, the proposed method shows superior performance around edges with contours better localized.

Fig. 2. Reconstruction of PET images using structural information from CT: (left) wavelet-based denoising approach, (right) our STV-prior denoising and deconvolution method.


Relevant literature:

  • Federico Turkheimer, Nicolas Boussion, Alexander Anderson, Nicola Pavese, Paola Piccini, and Dimitris Visvikis, "Pet image denoising using a synergistic multiresolution analysis of structural (MRI/CT) and functional datasets.," J Nucl Med, vol. 49, no. 4, pp. 657–666, Apr 2008.

Duration: from 2008
Funding: none
Contact person: Filip Šroubek
Involved people: J. Boldyš, B. Zitová, M. Šorel
Involved extern: J. Šroubek, O. Bělohlávek (Na Homolce hospital)


  • F. Šroubek et al.,"PET image reconstruction using prior information fromCT or MRI", submitted to ICIP'09.