Investigating a novel split‐filter dual‐energy CT technique for improving pancreas tumor visibility for radiation therapy

Abstract Purpose Tumor delineation using conventional CT images can be a challenge for pancreatic adenocarcinoma where contrast between the tumor and surrounding healthy tissue is low. This work investigates the ability of a split‐filter dual‐energy CT (DECT) system to improve pancreatic tumor contrast and contrast‐to‐noise ratio (CNR) for radiation therapy treatment planning. Materials and methods Multiphasic scans of 20 pancreatic tumors were acquired using a split‐filter DECT technique with iodinated contrast medium, OMNIPAQUE TM. Analysis was performed on the pancreatic and portal venous phases for several types of DECT images. Pancreatic gross target volume (GTV) contrast and CNR were calculated and analyzed from mixed 120 kVp‐equivalent images and virtual monoenergetic images (VMI) at 57 and 40 keV. The role of iterative reconstruction on DECT images was also investigated. Paired t‐tests were used to assess the difference in GTV contrast and CNR among the different images. Results The VMIs at 40 keV had a 110% greater image noise compared to the mixed 120 kVp‐equivalent images (P < 0.0001). VMIs at 40 keV increased GTV contrast from 15.9 ± 19.9 HU to 93.7 ± 49.6 HU and CNR from 1.37 ± 2.05 to 3.86 ± 2.78 in comparison to the mixed 120 kVp‐equivalent images. The iterative reconstruction algorithm investigated decreased noise in the VMIs by about 20% and improved CNR by about 30%. Conclusions Pancreatic tumor contrast and CNR were significantly improved using VMIs reconstructed from the split‐filter DECT technique, and the use of iterative reconstruction further improved CNR. This gain in tumor contrast may lead to more accurate tumor delineation for radiation therapy treatment planning.


| INTRODUCTION
Pancreatic adenocarcinoma is the fourth-leading cause of cancer death in the United States. 8,18 Although surgery is the only established curative treatment option, 80% of patients with pancreatic cancer are not surgical candidates. Radiation therapy offers a local treatment option with recent evidence suggesting that accurately focused dose-escalated radiation therapy may increase median survival and potential for surgical resection. 3,11 However, radiation therapy for pancreatic adenocarcinomas can be a challenge because they have poor innate contrast compared to surrounding healthy pancreatic tissue. 12,13,15 For radiation therapy treatment planning, the lack of tumor contrast makes it difficult to confidently delineate the target with conventional singleenergy computed tomography (SECT) even with iodine contrast. 20 Accurate tumor delineation is crucial for successful radiation therapy, 14 particularly in the pancreas where other radiation-sensitive organs are in close proximity. Recent work has shown that dose escalation can increase survival for pancreatic patients, further increasing the need to clearly visualize and accurately delineate the tumor for treatment planning. 7,17 Fortunately, recent efforts have been dedicated to using dualenergy computed tomography (DECT) as an optimal CT modality to increase the detectability of pancreatic tumors. 4,6,[8][9][10]12,16 DECT is an imaging modality that utilizes two different photon spectra to image patient anatomy. Since DECT provides information about the attenuation properties of tissues at two different energies, tissues with similar density but different elemental composition can be differentiated. 2 DECT images have significant advantages over conventional SECT specifically when imaging the abdomen; DECT applications in the abdomen include, but are not limited to, depicting small liver lesions, differentiating renal masses, and improving depiction of pancreas tumors. 1 Several studies have been published that investigate the use of DECT techniques for improving pancreas tumor contrast. In these studies, DECT offered improvements in tumor conspicuity and delineation compared to conventional 120 kVp CT. 1,4,8,10,16 A novel technique for single-source DECT was recently introduced as an additive feature to the Siemens SOMATOM Definition Edge CT scanner (Siemens Healthcare, Forchheim, Germany). The SOMATOM Definition Edge is now available with a removable gold and tin split-filter for DECT acquisition, known as TwinBeam (TwinBeam Dual Energy; Siemens). This system may offer a cost-effective solution for DECT applications in radiation therapy. The TwinBeam system is an innovative DECT modality that utilizes a split-filter to spatially separate a helical 120 kVp x-ray beam into a low-and high-energy beam along the longitudinal axis. TwinBeam allows for the low-and highenergy data of the same location in the patient to be acquired within two tube rotations. The temporal coherence between the low-and high-energy acquisition gives TwinBeam the capability to image dynamic contrast, making this modality a candidate for DECT imaging of pancreatic adenocarcinoma where iodine contrast is needed to differentiate healthy pancreatic tissue from tumor. However, in comparison to other dual-energy techniques that utilize a low-energy 80 kVp beam and high-energy 140 kVp beam, the split-filter technique of TwinBeam has inferior spectral separation. 2,5 The effects of this limited spectral separation on image quality, specifically in the pancreas have yet to be investigated.
Although some studies have investigated the image quality of   TwinBeam DECT scans, 2,5 none have investigated its utility for radiation therapy treatment planning, nor have any studies investigated the   use of TwinBeam for identifying and delineating pancreatic tumors. This work investigates tumor contrast while considering the noise characteristics by calculating tumor contrast-to-noise ratios (CNR).
CNR offers a more comprehensive view of image quality as both contrast and noise play a role in tumor segmentation during the treatment planning process. To the author's knowledge, there has not been any study to date that investigates the contrast between healthy pancreatic parenchyma and the entire gross target volume (GTV), rather a selected subsection of the GTV through a small region of interest (ROI) within the tumor. The goal of this work is to quantitatively determine if the split-filter DECT technique of TwinBeam can improve the contrast and the CNR of pancreatic GTVs with the long-term goal of improving tumor delineation for radiation therapy treatment. Due to the added beam filtration of the TwinBeam system, roughly two-thirds of the x-ray beam is filtered prior to reaching the patient. As a result, large tube currents are required to achieve CTDI vol similar to conventional SECT acquisitions. Due to tube current limitations, patient size, and scan length requirements for individual patients, the imaging protocol varied from patient to patient.

2.A | Patients and CT simulation
The machine effective mAs ranged between 1350 and 1500 mAs.
The automatic tube current modulation was not used, and the CTDI vol ranged from 21.6 to 33.6 mGy. Images were acquired with a pitch of 0.3 to 0.45, a rotation time of either 0.5 or 1 s per rotation and reconstructed at a slice thickness of 3 mm.

2.B | Image reconstruction
The Siemens syngo.via software was used to reconstruct virtual monoenergetic images, called Monoenergetic Plus images, as well as images that mimic the appearance of conventional single-energy 120 kVp images, called mixed images. A mixed image is a weighted sum with of the low-and high-energy datasets to create an image with HU values equivalent to a SECT image at 120 kVp and is therefore referred to as a 120 kVp-equivalent image. On the other hand, the virtual monoenergetic images (VMIs) used in this study depict how an object would appear if it was imaged using a monoenergetic x-ray source and are reconstructed using a novel monoenergetic algorithm (nMERA). 6 The possible reconstructed energies for a VMI range from 40 to 190 keV. For this study, VMIs were reconstructed at energies from 40 to 90 keV at 5 keV increments to investigate the change in contrast and CNR as a function of energy. Based on this preliminary analysis, the remainder of our study focused on two VMI energies: 40 and 57 keV. The VMI at 40 keV was chosen because it demonstrated the greatest CNR for pancreatic tumors, and the VMI at 57 keV was chosen based on physician initial preference.
Due to the increase in noise for low-energy VMIs, the role of iterative reconstruction on DECT images was investigated. In addition to filtered back projection (FBP) with the D30 reconstruction kernel, the latest generation of Siemens iterative reconstruction called Advanced Modeled Iterative Reconstruction (ADMIRE; Siemens Healthcare, Forchheim, Germany) using the Q30 reconstruction kernel was also used. ADMIRE is a model-based iterative reconstruction algorithm designed to decrease noise as well as metal and cone beam image artifacts by analyzing the data in both the Fourier and image domain. 6,19 ADMIRE was applied to the low-and high-energy datasets individually at a strength of 2 (ADMIRE 2) out of a maximum strength of 5; level 2 represents a low to medium level of noise suppression due to iterative reconstruction.
In summary, two raw datasets were acquired for each patient: a pancreatic contrast phase and a portal venous contrast phase. For each contrast phase, the raw data was reconstructed using two reconstruction methods, FBP and ADMIRE 2. For each reconstruction method, three dual-energy images were generated: a mixed 120 kVpequivalent image, a VMI at 57 keV image, and a VMI at 40 keV image ( Fig. 1).

2.C | Contrast and contrast-to-noise ratio analysis
All dual-energy images were evaluated using the MIMvista software (MIM Software Inc. Cleveland, OH). Three ROIs were created to evaluate tumor contrast and CNR. This study assessed the whole GTV, as defined by an experienced radiation oncologist on the pancreatic phase VMIs at 57 keV. The attenuation of healthy pancreatic parenchyma was measured using an ROI placed near the GTV within the pancreas that avoided stents, macroscopic vessels, and the pancreatic duct. The placement of the GTV and pancreatic parenchyma ROI contours for a single patient is shown in Fig. 2

| RESULTS
To determine the reconstruction energy for the VMIs that produced the greatest GTV contrast and CNR, VMIs were reconstructed at energies ranging from 40 to 90 keV at 5 keV increments (Fig. 3).
Among all patients, the reconstruction energy at 40 keV produced the greatest contrast and CNR. Figure 4 shows the GTV contrast, noise and CNR from VMIs at energies ranging from 40 to 90 keV reconstructed from FBP pancreatic phase data. The remainder of our results focus on comparing VMIs at 40 and 57 keV against mixed 120 kVp-equivalent images, which were used to represent conventional SECT images.

3.A | Contrast
The mean contrast values for the mixed 120 kVp-equivalent images, VMIs at 57 keV, and VMIs at 40 keV for both contrast phases are shown in Fig. 5(a). The mean ± standard deviation (SD) GTV contrast for the pancreatic phase datasets using FBP was 15.9 ± 19.9 HU for the mixed 120 kVp-equivalent images. The VMIs at 57 keV increased the GTV contrast to 40.7 ± 27.7 HU, which represents a 219% increase in contrast (P = 0.0025). The VMIs at 40 keV increased the GTV contrast to 93.7 ± 49.6 HU for a mean contrast improvement of 665% compared to the mixed 120 kVp-equivalent images (P < 0.0001). The mean ± SD GTV contrast for the portal venous phase datasets using FBP was 6.01 ± 15.2 HU, 16.4 ± 20.9 HU, and 41.5 ± 34.9 HU for the mixed 120 kVp-equivalent images, VMIs at 57 keV, and the VMIs at 40 keV, respectively (P < 0.0001). The GTV contrast was greater for all pancreatic phase images when compared to the portal venous phase images (P < 0.0001). On average, images reconstructed with ADMIRE had slightly greater contrast but this improvement was statistically insignificant (P = 0.8717). The mean ± SD GTV contrast for the pancreatic and portal venous phase datasets reconstructed with the FBP or ADMIRE and P-values are displayed in Table 1.

3.B | Noise
The mean ± SD image noise of the mixed 120 kVp-equivalent images, VMIs at 57 keV, and VMIs at 40 keV with the FBP was 13.2 ± 4.38 HU, 20.3 ± 8.45 HU, and 28.0 ± 12.5 HU, respectively, averaged over the pancreatic and portal venous phase datasets (P < 0.0001). There was no difference in image noise between the two contrast phases (P = 0.919). The image noise was 52% higher for the VMIs at 57 keV and 110% higher for VMIs at 40 keV compared to the mixed 120 kVp-equivalent images. ADMIRE 2 F I G . 2. VMI at 57 keV with the pancreatic adenocarcinoma GTV contour in red and the normal pancreas tissue contour in blue.

3.C | Contrast-to-noise ratio
The mean CNR for the mixed 120 kVp-equivalent images and the VMIs reconstructed at 57 and 40 keV for both the pancreatic and portal venous phases are shown in Fig. 5

| DISCUSSION
In this study, TwinBeam was investigated to improve the delineation of pancreatic adenocarcinoma for radiation therapy treatment planning. VMIs acquired using TwinBeam were compared against mixed 120 kVp-equivalent images, which served as a baseline since these images represent conventional single-energy CT images. Entire  The reconstruction algorithm used to create the VMIs from Twin-Beam data is a novel monoenergetic reconstruction algorithm (nMERA) that performs regional spatial frequency-based recombination of high attenuation at low photon energy images and lower image noise at higher photon energies to obtain the best possible image noise level. This algorithm is different than the standard monoenergetic reconstruction algorithm (sMERA). 6 The difference in noise characteristics between the DECT systems results in different VMI energies producing the greatest CNR. The VMI energy that produced the greatest CNR was 40 keV for the TwinBeam system, while the optimal energy varied by patient (52.5 ± 7.7 keV) for the fast-kVp switching technique. 16  dual-source systems. This comparison highlights that in addition to the spectral separation between different dual-energy CT systems, the gain in CNR also depends on the algorithm used to generate monoenergetic images.
TwinBeam is a new single-source DECT, which may aid in tumor delineation for radiation therapy treatment planning. TwinBeam shows promise for DECT simulations, which aim to capture a dynamic bolus of contrast. This work demonstrates that VMIs reconstructed using the TwinBeam system provide greater CNR between pancreatic tumors and healthy pancreatic parenchyma than virtual single-energy CT images. For pancreatic tumors which are historically difficult to differentiate, this increase in CNR may increase the ability to accurately segment these tumors for radiation therapy treatment planning, which has the potential to lead to more effective radiation therapy treatment. However, definitive improvements in tumor delineation cannot be stated without further investigation of pancreas GTV segmentation reproducibility and accuracy.

ACKNOWLEDGMENTS
The authors thank the students and staff of the University of Wisconsin (UW) Medical Radiation Research Center and the costumers of the UW Accredited Dosimetry Laboratory, whose calibrations helped fund this research. The authors also acknowledge Dr. Nilesh Mistry from Siemens Healthineers for his valuable comments on this work.

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