Material Decomposition

In Figure 12.10 we find material-specific images that were extracted from the spectral CT-scans shown in Figure 12.9. Material decomposition was performed in the CT domain using a linear model, that is, the material-specific images are linear combinations of the multi-energy images. The model was trained using a least squares regression to recover the ground truth [39].

Inspecting Figure 12.10, it is not hard to again find a difference in noise when comparing SPM with CSM. the former showing a comparably strong noise component that is visible for each of the three materials. In particular, the difference in density between the PMMA phantom material (1.18 g/cm3) and the water contained within the contrast agent solutions (1 g/cm3) is barely visible here. Charge summing, on the contrary, resolves this density contrast much better. It also reveals the existence of the lower of the two gadolinium concentrations, whose existence is mostly concealed within the SPM scan.

Material-specific images extracted from the two spectral CT scans shown in Figure 12.9 [39]. SPM

FIGURE 12.10 Material-specific images extracted from the two spectral CT scans shown in Figure 12.9 [39]. SPM: single pixel mode; CSM: charge summing mode.

The attentive reader who closely examines these material-specific CT slices will also identify another important difference between the two modes of operation: The SPM measurement is notably sharper than its CSM counterpart, which is best visible when comparing the two iodine images. It is tempting to blame the threshold dependent PSF as the origin of this behavior, and certainly this effect contributes to some extent. However, charge summing introduces yet another source of blurring to a photon counter: If the energy of a secondary photon is larger than the energy deposited at the original site of interaction, a hit will be assigned to where the secondary photon w'as reabsorbed. This can be directly concluded from the w orking principle of charge summing, which ensures to assign a hit to the pixel receiving most of the charge. While this effect does not increase image noise, it represents a source of blurring especially for characteristic x-rays, w'hich are emitted isotropically within the CdTe sensor under study, and w'ith a yield of more than 80%. Hence, the only w'ay to efficiently suppress this blurring is to use photon energies that are at least twice the energy of the largest fluorescence energy occurring.

< Prev   CONTENTS   Source   Next >