2.1

Data acquisition from a dual-layer flat panel detector prototype

With a dual-layer FPD, dual-energy data can be acquired simultaneously. As illustrated in Fig. 1, a higher energy spectrum is obtained from the bottom layer which is essentially filtered by the top layer detector. Mathematically, low- and high-energy projections can be expressed as

Figure 1.

Left: The detailed structure of a prototype dual-layer FPD used in our study.Right: the normalized effective spectrum for top and bottom layer.

where, S(E) denotes the incident spectrum of source; and denote the detector responses of top layer and bottom layer, respectively; denotes the attenuation of top-layer material and possible inter-layer filter; S(E) · η_T(E) and therefore indicate the overall detective spectra of low- and high-energy scan, respectively.

2.2

Reconstruction from combined dual-layer projections

On a dual-layer flat panel detector, the top layer is more likely to absorb low-energy x-ray photons, while relatively more high-energy x-ray photons are absorbed in the bottom layer. Naturally, by weighting two layers of projection data properly, noise reduction in the corresponding reconstruction will be achieved as more signals (x-ray photons) will be utilized. An optimal weight factor ω can be determined by minimizing the noise on reconstructed images from combined dual-layer projections as below.

2.3

Physics model-guided material decomposition

First, conventional head CT images with decreased noise are reconstructed from adaptively combined dual-layer projection data, followed by a dual-layer multi-material spectral correction (dMMSC) to generate beam hardening free images.⁵ After a regular projection-domain material decomposition (MD) using dual-layer detector data that is usually quite sensitive to low-signal x-ray photon and energy separation, the dMMSC corrected projections are used as a physics-model based guidance to further enhance the dual-layer MD performance.

2.3.2

Dual-layer multi-material spectral correction

As we know, due to the significant bony structures in human head, a water correction is not enough to remove beam hardening artifact in head CT images. In the literature, multi-material spectral correction has been developed as a poster-reconstruction method for conventional CT scan to estimate and correct for bone’s beam hardening impact. In this work, we extend this kind of method to dual-layer projection data, which consists of the following steps. First, an initial reconstruction after water correction is conducted to estimate the distribution of bony structures that can be easily segmented out from soft tissues. Then, projection of bony structures P_b can be computed using forward projection. With the help of the bone projection, a multi-material spectra-corrected projection that is beam-hardening free can be computed as

where, f(P_t, P_b) represents the bone-induced beam-hardening error and can be modeled and calculated in advance by simulations or measurements.

2.3.3

material decomposition enhancement using dMMSC

In order to improve the dual-energy performance from the dual-layer FPD, beam hardening free projections after dMMSC is adopted as a guidance to minimize the material decomposition errors and generate a high-quality head image. An optimal hybrid VM projections can be generated using an adaptive fusion of projection with dMMSC and VM projection at an efficient energy as

where, P_dMMSC and denote the spectra-corrected projection and VM projection, respectively.Basis material projections can also be re-fined by using the hybrid VM projection at optimal keV and a VM projection generated at higher keV .

3.1

Experimental system set up

Our study is based on a benchtop CBCT system equiped with a prototype dual-layer FPD in our lab.The prototype dual layer FPD (Varex imaging, Salt Lake City, UT, USA) consists of two amorphous silicon (a- Si) panels with pixel size of 150 μm, both deposited with a 550 μm-thick CsI scintillator, with no additional intermediate filter placed in between [Fig. 2].The source-to-axis distance was set to 750mm and the source-to-detector distance to 1184 mm. In our study, 720 projections over 360 degree were collected at 30 fps in a 3×3 binning mode. The X-ray source (Varex Imaging G-242, Salt Lake City, UT, USA) at 120 kV, 64 mA and 5 ms of pulse width. In our current study, a narrowed X-ray collimation was implemented to get rid of X-ray scatter’s impact which is another big challenge to CBCT spectral imaging but is under a separate investigation. In our study, a 120 kV X-ray source is used. The estimated effective spectrum for the top and bottom layer is shown in Fig. 1 and the average energy separation between low- and high-energy spectra can reach a level of 17 keV without object in the beam.

Figure 2.

The benchtop CBCT system using DL FPD.

3.2

Combined reconstruction from dual-energy projections

By combining the dual layer projection data using different weighting factor, an optimal weighting factor can be determined empirically by measuring the standard deviations of selected ROIs. As shown in Fig. 3, the measurements of signal to noise ratio (SNR) on reconstructed images from top-layer, bottom-layer and combined data are shown beside the selected ROIs. Our preliminary results suggest that the standard deviation can be reduced by roughly 10% when compared with that of the top layer alone.

Figure 3.

Reconstruction from low- and high-energy projections and our proposed weighted combination of the projections. Display window: [-100, 100] HU (Hounsfield Units).

3.3

Physics model-guided material decomposition

The results of basis material decomposition of a head phantom are shown in Fig. 4. The VM images at 63 keV was generated and it is seen that most beam hardening artifacts caused by the bony objects are removed. However, in some locations strong streak artifacts occur badly in the CT images, which can be well suppressed by using our proposed approach. Firstly, we use the spectra-corrected projection at 63 keV as a guidance of VM projection at the same keV to generate an improved VM projection which is combined with generated VM image at 90 keV to refine the water and bone image. By using our physics based material decomposition, we can see that most streak artifacts can be removed in the basis material images and VM images.

Figure 4.

Results of first-pass material decomposition (1st and 3rd rows) and physics model-guided material decomposition (2nd and 4th rows) at two head phantom sections. Display windows for the water, bone and VM images are [-500, 2000] mg/ml, [0, 2000] mg/ml, [-150, 150] HU, respectively.