PET Attenuation Map Generation for Brain Imaging from Virtual Non-Contrast and Virtual Non-Iodine Images Obtained with Photon Counting CT Scanners

Abstract

This study presents a transformation scheme for converting measured Hounsfield units (HUs) into linear attenuation coefficients (LACs) for 511 keV gamma photons in iodine contrast-enhanced angiographic CT images. The method is based on spectral CT data acquired from tissue-equivalent inserts of the Sun Nuclear electron density phantom. The resulting transformation will be used to develop an attenuation map template for our in-house developed BrainPET 7T system. Three approaches were used to obtain the attenuation maps. The first approach is based on angiographic spectral CT scans with iodinated contrast agent administration, and the computation of virtual non contrast (VNC) and virtual non-iodine (VNI) images. In this case, two images are combined: bone regions are segmented from the VNI images, and the corresponding voxel values in the VNC images are replaced, yielding a single quantitative monochromatic CT image. The second approach relies exclusively on VNC images, whereas the third approach uses only VNI images. The transformation schemes from HUs to LACs were established for different tube voltages (120 kV and 140 kV) and virtual monochromatic energies (64 keV, 67 keV, 70 keV, 76 keV, 80 keV, and 100 keV), showing the HUs dependency on the energy setting of the reconstructed virtual monochromatic energies. The accuracy and precision of the proposed transformation scheme were further evaluated using previously acquired angiographic CT datasets. Spectral CT can be expected to minimize the systematic error when converting HUs to LACs for 511 keV gamma rays using virtual monochromatic images (VMIs).