Segmented attenuation correction in whole body PET using neighbourhood EM clustering

Attenuation correction of positron emission data provides an enhanced image interpretation by correcting for artifacts generated by the photon attenuation process. The duration of transmission scans is a serious limitation for whole body PET scanning, which can involve up to 10 separate bed positions, for both emission and transmission scans. In order to limit the overall scan time, shorter duration transmission scans can be acquired. However, low statistic transmission scans introduce noise into attenuation corrected PET images. Segmentation of transmission scans into a small number of attenuation coefficient classes is frequently applied to reduce noise in the calculated attenuation coefficient. Many of the segmentation algorithms applied in PET use intensity information only and ignore or make only limited use of spatial clustering. A novel segmentation algorithm, that combines the Expectation Maximization (EM) algorithm with local spatial information (within 2D slices) to produce a fuzzy clustering, has been applied to the segmentation of PET transmission scans. Qualitative and quantitative results from phantom and clinical studies are presented here.