Figure 4

Reconstructed tomograms from the PET2 data set using three different reconstruction methods based on 30% sampling. The PET2 data set was taken from a 53 year-old man who sought medical attention due to a grand mal seizure, with a brain biopsy revealing grade IV astrocytoma. As with the reconstructed tomograms from the PET1 data set, it can be observed that the FBP method results in the most significant artifacts in both the background and imaged brain areas when compared to TVR and HNLR, the TVR method results in reduced artifacts but noticeable loss in fine detail in the imaged brain area, and the HNLR method results in the least amount of artifacts and better preservation of fine detail in the imaged brain area. It is very clear that streaking artifacts (caused by sampling at a sub-Nyquist sampling fraction) is prominently overlaid on the entire brain region in the reconstruction produced using the FBP method (see red arrows), while they are significantly reduced in the reconstructions produced by TVR and HNLR. It is also important to highlight that, within the same brain region in the HNLR reconstruction, there is an overall improvement in sharpness when compared to the TVR reconstruction (see green arrows). Two line profiles (indicated by orange lines) from the PET data sets reconstructed using the methods are shown at the bottom, and further illustrates the reduction in artifacts and improved contrast achieved by TVR and HNLR when compared to FBP.