Investigating split‐filter dual‐energy CT for improving liver tumor visibility for radiation therapy

Abstract Purpose Accurate liver tumor delineation is crucial for radiation therapy, but liver tumor volumes are difficult to visualize with conventional single‐energy CT. This work investigates the use of split‐filter dual‐energy CT (DECT) for liver tumor visibility by quantifying contrast and contrast‐to‐noise ratio (CNR). Methods Split‐filter DECT contrast‐enhanced scans of 20 liver tumors including cholangiocarcinomas, hepatocellular carcinomas, and liver metastases were acquired. Analysis was performed on the arterial and venous phases of mixed 120 kVp‐equivalent images and VMIs at 57 keV and 40 keV gross target volume (GTV) contrast and CNR were calculated. Results For the arterial phase, liver GTV contrast was 12.1 ± 10.0 HU and 43.1 ± 32.3 HU (P < 0.001) for the mixed images and 40 keV VMIs. Image noise increased on average by 116% for the 40 keV VMIs compared to the mixed images. The average CNR did not change significantly (1.6 ± 1.5, 1.7 ± 1.4, 2.4 ± 1.7 for the mixed, 57 keV and 40 keV VMIs (P > 0.141)). For individual cases, however, CNR increases of up to 607% were measured for the 40 keV VMIs compared to the mixed image. Venous phase 40 keV VMIs demonstrated an average increase of 35.4 HU in GTV contrast and 121% increase in image noise. Average CNR values were also not statistically different, but for individual cases CNR increases of up to 554% were measured for the 40 keV VMIs compared to the mixed image. Conclusions Liver tumor contrast was significantly improved using split‐filter DECT 40 keV VMIs compared to mixed images. On average, there was no statistical difference in CNR between the mixed images and VMIs, but for individual cases, CNR was greatly increased for the 57 keV and 40 keV VMIs. Therefore, although not universally successful for our patient cohort, split‐filter DECT VMIs may provide substantial gains in tumor visibility of certain liver cases for radiation therapy treatment planning.

Liver cancer is one of the leading causes of cancer-related deaths in the United States. Unfortunately, at the time of diagnosis, the majority of cases are advanced and therefore not candidates for curative treatment. Surgical resection is the established curative treatment but because of the extent of the majority of liver tumors and their venous involvement at diagnosis, they are unresectable. Radiation therapy is the most common localized treatment option for unresectable liver cancers, and recent studies have shown that doseescalated stereotactic body radiotherapy (SBRT) improves local control and may decrease tumor size for resection. 1 However, doseescalated SBRT requires precision targeting which can be challenging due to inaccurate liver tumor delineation on conventional computed tomography (CT) images. 1 Several studies have demonstrated that dual-energy CT (DECT) greatly improves the delineation and conspicuity of liver tumors. 2,3 Dual-energy CT is the acquisition of two 3-dimensional attenuation datasets using both low-and high-energy photon spectra during a single CT protocol. The low-and high-energy spectra are commonly achieved through fast kVp switching, two sequential scans, dual-layer detector, or using two x-ray sources. The low-and highenergy spectra are also achievable by placing filters within the beam to alter the mean energy of the spectra. DECT allows for the differentiation of tissues with the same density but different elemental composition and therefore has significant advantages over conventional SECT, specifically when imaging the abdomen. 2 When imaging the liver, low-energy images created from sequential scanning and fast kVp-switching DECT have been shown to increase iodine conspicuity and increase contrast of hypervascular liver tumors, including hepatocellular carcinomas (HCC) and metastases. [3][4][5][6] When considering the type of DECT modality, greater spectral separation results in better tissue differentiation, and greater temporal coherence between the low-and high-energy acquisitions reduces the impact of artifacts for dynamic contrast imaging. In addition to the previously mentioned techniques, single-source DECT can also be achieved using a split-filter technique available on the Siemens SOMATOM Definition Edge CT scanner (Siemens Healthcare, Forchheim, Germany). The Edge has an acquisition technique known as TwinBeam which introduces a gold and tin split filter for DECT acquisition. TwinBeam is a cost-effective and innovative DECT system. Although TwinBeam has a smaller time interval between the acquisition of the low-and high-energy data than sequential dual-energy scanning, it is more susceptible to artifacts when compared to fast-kVp switching, dual-source, and dual-layer detector DECT. The low-and high-energy datasets are acquired within two tube rotations making this modality applicable for dynamic contrast imaging.
Therefore, TwinBeam may also be beneficial for abdominal cancer imaging since studies have shown that two-phase imaging increases the detection of liver tumors. 7 However, a disadvantage of Twin-Beam is a lower spectral separation and, consequently, an inferior ability to differentiate tissues in comparison to other DECT techniques. 3,8,9 DECT techniques that utilize a low-energy 80 kVp and high-energy 140 kVp beams have been shown to increase liver tumor detection, but there has not been any study investigating the benefits of TwinBeam DECT on liver tumor delineation for radiation therapy applications. Much like a recent study that investigated the delineation of pancreas tumors using TwinBeam, this work investigates the gross target volume (GTV) contrast and contrast-to-noise ratio (CNR). 10 This work investigates several types of liver tumors, unlike recent DECT studies that have solely investigated hypo-or hypervascular liver tumors. 1,3,4,6,7 Liver metastases, hepatocellular carcinomas (HCC), and cholangiocarcinomas are all included for investigation in this work. The goal of this work is to quantitatively determine if TwinBeam DECT can improve the contrast and CNR of liver tumors in comparison to conventional single-energy CT imaging methods, with the long-term goal of improving the delineation of these tumors for radiation therapy treatment planning purposes.

2.A | Patient population and CT simulation
Patient information was collected for this study after Institutional Review Board approval for patients who received dual-energy imaging at CT simulation for radiation therapy at our institution between June 2016 and August 2018. Of the 20 patients with liver cancer who were included in this study, 14 were men and 9 were women.
The mean ± SD (range) of age was 67.1 ± 10 (39-83) years. The mean ± SD (range) of weight was 82.5 ± 12 (56.8-107.9) kg. On the basis of either histopathologic analysis or imaging follow up, 6 tumors were diagnosed as intrahepatic cholangiocarcinoma, 10 as metastatic liver cancer, and 4 as hepatocellular carcinoma. The study population included Stage I-IV liver cancer and Stage IV cancer of the esophagus, colon, and rectum that metastasized to the liver. The longest tumor dimension ranged from 1 to 14 cm.
The image acquisition has been thoroughly described in a previous study investigating pancreas tumors. 10 For this study, all patients were imaged with a dual-phase imaging protocol. The arterial and portal venous phase scans were acquired using patient-specific delays based on automatic bolus tracking of the abdominal aorta. cushions were used as immobilization devices. All patients were scanned with a dual-phase imaging protocol with the same amount of OMNIPAQUE™ IV nonionized iodine contrast medium regardless of patient weight. A bolus tracking technique was used to achieve the appropriate delay times.

2.B | Image reconstruction
Each raw dataset was reconstructed using the Siemens' iterative reconstruction algorithm, ADMIRE, at a strength of 2 of 5. ADMIRE 2 was used because it has shown to decrease image noise by 20% and is preferred strength by clinicians. 10 A mixed 120 kVp-equivalent image, a virtual monoenergetic image (VMI) at 57 keV, and a VMI at 40 keV were then generated for each phase, for a total of 6 different image sets for each liver tumor case (Fig. 1). The VMIs at 57 keV and 40 keV were generated using the Siemens Monoenergetic + application.

2.C | Contrast and contrast-to-noise ratio
The MIMvista software (MIM Software Inc. Cleveland, OH, USA) was used to analyze each image. The entire liver GTV was segmented by an experienced radiation oncologist on the arterial phase VMI at 57 keV, similar to what is done for radiation treatment planning at our institution. To investigate liver tumor GTV contrast, the surrounding healthy liver tissue was assessed using a nearby region of interest (ROI) placed within a homogenous region of healthy liver parenchyma avoiding any vessels and bile ducts. A 10 mm 2 ROI placed within a homogeneous region of the erector spinae muscle was used to assess image noise.
Gross target volume contrast was calculated as the absolute difference in HU between the healthy liver parenchyma and GTV, GTV contrast ¼ HU liver À HU GTV . GTV contrast was divided by the standard deviation of the ROI located in the erector spinae muscle to calculate GTV CNR, GTV CNR ¼ HU liver ÀHUGTV j j σ . The absolute contrast difference was used to calculate GTV contrast and CNR in order to analyze hypoattenuating and hyperattenuating liver tumors using the same methodology.
MATLAB was used for all statistical analyses. The difference in absolute GTV contrast and CNR between the mixed 120 kVp-equivalent images and VMIs was analyzed using paired t-tests. Statistical significance was determined using a P-value less than 0.05.

3.B | Noise
There was no statistical difference in noise between the arterial and venous phase across all images (P> 0.05). The mean ± SD image noise of the mixed 120 kVp-equivalent images, VMI at 57 keV, and VMIs at 40 keV was 8.1 ± 1.6 HU, 12.7 ± 2.0 HU, and 18.0 ± 2.7 HU, respectively (P < 0.001). Image noise was about 50% higher for the VMIs at 57 keV and 120% higher for the VMIs at 40 keV compared to the mixed 120 kVp-equivalent images.  can be very heterogeneous due to vascular heterogeneity causing regions of hypoxia or regions of greater enhancement. These hypoor hyperintense regions will then get averaged out when considering the GTV as a whole. Figure 4 shows examples of heterogeneous tumors from our study. For these specific cases, the VMIs at 40 keV did not have a greater GTV CNR than the mixed 120 kVp-equivalent images. We did, however, separate the cohort based on tumors that were not visually heterogeneous, and although the sample size was small, we did see significant improvements in CNR with the VMI at 40 keV for these specific cases. This included both hypo-and hyperattenuating tumors.
The heterogeneity of the GTV and the use of a small ROI were further investigated for Case 5. Figure 5 shows the histograms of the GTV of Case 5. These histograms provide a quantitative depiction of the heterogeneity of the tumor and one can conclude that an ROI placed in a high-contrast region will provide different CNR values than the average value of the entire GTV.

CONF LICT OF I NTERESTS
This work was partially funded by a collaboration with Siemens Healthineers.