부산대학교 도서관

Diagnosis , staging and treatment of disease depends on the morphological and functional information obtained from multimodality molecular imaging systems. The combination of functional and morphological information is now routinely performed to overcome the limitations of each individual modality. Attenuation of photons in the object under study is one of the main limitations of quantitative PET imaging. Attenuation correction plays a pivotal role in PET imaging. However, the availability of CT data on hybrid PET/CT scanners made it possible to build an accurate attenuation map. One of the well-known methods for generation of the attenuation map on PE/MRI systems is MR-based attenuation correction (MRAC) where image segmentation is used to classify MRI into several classes corresponding to different attenuation factors. In this study we investigate the effect of using different numbers of classes for the generation of attenuation maps on the accuracy of attenuation correction of PET data. The study was carried out using simulations of the XCAT phantom and 10 clinical studies. For the later, CT and PET images of 10 patients were used with CT-based attenuation correction assumed as reference. MRI was classified into different classes to produce two, three and four-class attenuation maps using the ITK library. The relative error showed that the lower number of classes will increase the global error over 8%. The elimination of bony structures from the attenuation map will cause a local error over 3%. In clinical studies, SUV mean and SUV max were calculated for each AC method. The results seem to indicate an underestimation of 11% because of neglecting bone.