PET-enhanced liver segmentation for CT images from combined PET-CT scanners

The use of functional (PET) information from PET-CT scanners to assist liver segmentation in CT data has yet to be addressed. In this work we implement PET data enhanced liver segmentation with CT. We utilize the difference in FDG uptake between the liver and adjacent organs to separate the liver from these structures, which have similar intensities in low-contrast CT. The relatively high normal FDG uptake, and hence high SUV of liver metabolism, allows an accurate estimation for liver segmentation in CT images. By deformable registration, the PET ROIs are mapped onto the CT images for the initial liver segmentation in CT. To overcome the different intensity values of CT images from different patients or over multiple temporal imaging sessions, the initial liver region in CT images is used to establish the accurate threshold criteria for CT liver segmentation. To prevent the deformable model from leaking into the adjacent tissues and structures, the feature images are computed to exclude and disconnect neighboring organs and tissues from liver. Our experimental results in 12 clinical PET-CT studies suggest that our algorithm is robust when dealing with livers of different shapes and sizes and from a range of different patients.