Gsi viewer

Enhanced DE-CT imaging was performed using a dual-energy gemstone spectral CT scanner (Discovery CT 750 HD; GE Healthcare, Tokyo, Japan) and a fast kilovoltage (kV) switching method. Details of this imaging method have been previously reported [19 (link)]. Briefly, a non-ionic, low-osmolar contrast medium dose of 600 mg I/kg body weight, with an iodine content of 300 or 350 mgI/ml, was administered. The total amount of contrast medium was intravenously injected within 30 s. Scanning began 25 s after initiating the injection of contrast medium. Slices used for data analysis had a thickness of 0.63 mm.
All CT images were transferred to a workstation (GSI Viewer, GE Healthcare) and subjected to data analyses. Using a pulmonary window (window width, 1000 HU; window level, −700 HU), a region of interest (ROI) was set at the maximum cross-sectional diameter of the tumor to surround the entire tumor on the CT image; the image was subsequently converted to an iodine (water) image to obtain an AID.

All CT images were transferred to a workstation (GSI Viewer, GE Healthcare, USA) and were subjected to data analyses. The region of interest (ROI) area was set at the maximum cross-sectional diameter of the tumor, surrounding the whole tumor on the CT image using a pulmonary window (window width, 1000 HU; window level, −700 HU) and the image was converted to an iodine (water) image, as shown in Fig. 1. The average iodine density obtained by gemstone spectral imaging software was evaluated with regard to variables (diagnosis, histology, tumor diameter, and BED₁₀) and local control.
Fig. 1.

Location of the lung tumor regions of interest (red circle): computed tomography image of the pulmonary window (A); iodine (water) image (B).

An in vitro experiment was carried out to evaluate the accuracy of iodine concentration measurements. A set of 8 test tubes containing iodine at 0.5 –30.0 mg/mL were scanned by the SSDE CT protocol using fast tube voltage alternating between 80 and 140 kVp with collimation thickness of 0.625 mm, rotation speed of 0.6 s, and helical pitch of 0.983:1. The iodine concentrations were measured using a circular region of interest that covered 80% of the test tube's cross section on the iodine-based material decomposition images. Then, the iodine level measured in each test tube was compared with the known concentration. Finally, a radiologist (∗∗) undertook the quantitative measurements with dedicated imaging viewer (GSI Viewer; GE Healthcare) to analyze the material decomposition images.

The training labels were generated from the acquired human subject CT data: (1) 140-kV component of the DECT sinogram data was obtained using a proprietary software tool provided by the CT vendor. This single-kV sinogram was used as y. (2) Water and iodine basis images, namely a_w and a_I, and their forward-projections, namely P_w and P_I. The material basis image training labels were generated from a GSI DECT image post-processing software (GSI Viewer, GE Healthcare). (3) Effective energy labels (ε). The minimal-norm problem defined in Equation (7) was solved to obtain ε from y, P_w, and P_I. The same training data sets were used for both the module-wise pretraining and end-to-end training. Pretraining each module separately was to search for a solution quickly and to ensure the module approximately models the underlying imaging physics designed for the module. Then the end-to-end training pushed the Color-Resolving module towards a better solution using physics constraints imposed by modules 𝒞 and 𝒮. Figure 7 shows the observed convergence of the Color-Resolving module in the pretraining stage and end-to-end training stage using the same training datasets. During both training stages, the losses for the Color-Resolving module monotonically decreased, and the values were effectively reduced in the end-to-end training stage compared to the pretraining stage.