Comparative Assessment of Energy-Mapping Approaches in CT-Based Attenuation Correction for PET

Reliable quantification in positron emission tomography (PET) requires accurate attenuation correction of emission data, which in turn entails accurate determination of the attenuation map (µ-map) of the object under study. One of the main steps involved in CT-based attenuation correction (CTAC) is energy-mapping, or the conversion of linear attenuation coefficients (µ) calculated at the effective CT energy to those corresponding to 511 keV.

The aim of this study is to compare different energy-mapping techniques including scaling, segmentation, the hybrid method, the bilinear calibration curve technique and the dual-energy approach to generate the µ-maps required for attenuation correction. In addition, our newly proposed method involving a quadratic polynomial calibration curve was also assessed. The µ-maps generated for both phantom and clinical studies were assessed qualitatively and quantitatively. A cylindrical polyethylene phantom containing different concentrations of K₂HPO₄ in water was scanned and the µ-maps calculated from the corresponding CT images using the above-referenced energy-mapping methods. The CT images of five whole-body data sets acquired on a GE Discovery LS PET/CT scanner were employed to generate µ-maps using different energy-mapping approaches that were compared with the µ-maps generated at 511 keV using ⁶⁸Ge/⁶⁸Ga rod sources. In another experiment, the evaluation was performed on PET images of a clinical study corrected for attenuation using µ-maps generated using the above described methods. The evaluation was performed for three different tissue types, namely, soft tissue, lung, and bone.

All energy-mapping methods yielded almost similar results for soft tissues. The mean relative differences between scaling, segmentation, hybrid, bilinear, and quadratic polynomial calibration curve methods and the transmission scan serving as reference were 6.60%, 6.56%, 6.60%, 5.96%, and 7.36%, respectively. However, the scaling method produced the largest difference (16%) for bone tissues. For lung tissues, the segmentation method produced the largest difference (14.9%). The results for reconstructed PET images followed a similar trend. For soft tissues, all energy-mapping methods yield results in nearly the same range. However, in bone tissues, the scaling method resulted in considerable bias in the µ-maps and the reconstructed PET images. The segmentation method also produced noticeable bias especially in regions with variable densities such as the lung, since a single µ is assigned to the lungs. Apart from the aforementioned case, despite small differences in the generated µ-maps, the use of different energy-mapping methods does not affect, to a visible or measurable extent, the reconstructed PET images.