Purpose: Accurate attenuation correction is important for PET quantification. Often, a segmented attenuation map is used, especially in MRI-based attenuation correction. As deriving the attenuation map from MRI images is difficult, different errors can be present in the segmented attenuation map. The goal of this paper is to determine the effect of these errors on quantification.
Methods: The authors simulated the digital XCAT phantom using the GATE Monte Carlo simulation framework and a model of the Philips Gemini TF. A whole body scan was simulated, spanning an axial field of view of 70 cm. A total of fifteen lesions were placed in the lung, liver, spine, colon, prostate, and femur. The acquired data were reconstructed with a reference attenuation map and with different attenuation maps that were modified to reflect common segmentation errors. The quantitative difference between reconstructed images was evaluated.
Results: Segmentation into five tissue classes, namely cortical bone, spongeous bone, soft tissue, lung, and air yielded errors below 5%. Large errors were caused by ignoring lung tissue (up to 45%) or cortical bone (up to 17%). The interpatient variability of lung attenuation coefficients can lead to errors of 10% and more. Up to 20% tissue misclassification from bone to soft tissue yielded errors below 5%. The same applies for up to 10% misclassification from lung to air.
Conclusions: When using a segmented attenuation map, at least five different tissue types should be considered: cortical bone, spongeous bone, soft tissue, lung, and air. Furthermore, the interpatient variability of lung attenuation coefficients should be taken into account. Limited misclassification from bone to soft tissue and from lung to air is acceptable, as these do not lead to relevant errors.