Feasibility study of ultrasound imaging for stereotactic body radiation therapy with active breathing coordinator in pancreatic cancer

Abstract Purpose Stereotactic body radiation therapy (SBRT) allows for high radiation doses to be delivered to the pancreatic tumors with limited toxicity. Nevertheless, the respiratory motion of the pancreas introduces major uncertainty during SBRT. Ultrasound imaging is a non‐ionizing, non‐invasive, and real‐time technique for intrafraction monitoring. A configuration is not available to place the ultrasound probe during pancreas SBRT for monitoring. Methods and Materials An arm‐bridge system was designed and built. A CT scan of the bridge‐held ultrasound probe was acquired and fused to ten previously treated pancreatic SBRT patient CTs as virtual simulation CTs. Both step‐and‐shoot intensity‐modulated radiation therapy (IMRT) and volumetric‐modulated arc therapy (VMAT) planning were performed on virtual simulation CT. The accuracy of our tracking algorithm was evaluated by programmed motion phantom with simulated breath‐hold 3D movement. An IRB‐approved volunteer study was also performed to evaluate feasibility of system setup. Three healthy subjects underwent the same patient setup required for pancreas SBRT with active breath control (ABC). 4D ultrasound images were acquired for monitoring. Ten breath‐hold cycles were monitored for both phantom and volunteers. For the phantom study, the target motion tracked by ultrasound was compared with motion tracked by the infrared camera. For the volunteer study, the reproducibility of ABC breath‐hold was assessed. Results The volunteer study results showed that the arm‐bridge system allows placement of an ultrasound probe. The ultrasound monitoring showed less than 2 mm reproducibility of ABC breath‐hold in healthy volunteers. The phantom monitoring accuracy is 0.14 ± 0.08 mm, 0.04 ± 0.1 mm, and 0.25 ± 0.09 mm in three directions. On dosimetry part, 100% of virtual simulation plans passed protocol criteria. Conclusions Our ultrasound system can be potentially used for real‐time monitoring during pancreas SBRT without compromising planning quality. The phantom study showed high monitoring accuracy of the system, and the volunteer study showed feasibility of the clinical workflow.

study showed high monitoring accuracy of the system, and the volunteer study showed feasibility of the clinical workflow.

K E Y W O R D S
intrafraction, pancreas, SBRT, ultrasound

| INTRODUCTION
Pancreatic cancer remains one of the leading causes of cancer deaths in the United States. 1 Currently, the only curable treatment option has been surgical resection. However, most patients are unresectable, as only 20% of patients are surgical candidates. 2 For the unresectable patients, including the locally advanced and borderline resectable pancreatic cancer patients, the standard of care includes chemotherapy and radiation therapy. Although the optimal sequence, radiation technique, and total dose have not been well defined yet, recent advances in radiation therapy have improved the overall survival rate. 3 Our institution experience has previously been reported to utilize definitive five-fraction stereotactic body radiation therapy (SBRT) for locally advanced pancreatic cancer patients and borderline resectable pancreatic cancer patients. The report shows that chemotherapy of systemic gemcitabine followed by SBRT resulted in additional advancement toward optimizing patient outcomes. 4,5 Some patients even had margin-negative resection and complete pathologic response with no remaining cancer cells found at the time of surgery. 6 Despite our institutional pancreas SBRT experience, early radiation therapy studies likely had higher toxicity rates due to the lack of fractionation, inadequate motion management, lack of image guidance, and lack of specific dose constraints for organs at risk. The motion of the pancreas due to patient respiration is the primary source of intrafraction treatment uncertainties. 7 Commonly employed motion management techniques include respiratory gating, active breathing coordinator (ABC), and abdominal compression. 8 Due to the possibility that the stomach and duodenum may be pushed into the target volume, resulting in increased radiation toxicity to these structures, it is not recommended to use abdominal compression techniques. These methods physically restrict the abdominal muscle movement with either a plate or belt that applies a significant amount of pressure. In addition to motion management, intrafraction monitoring is becoming available in daily clinical use as it can verify the target location during the radiation therapy and thus eliminate the intrafraction treatment uncertainty due to motion, even under motion management techniques. [9][10][11] Currently, several intrafraction motion monitoring methods have been developed. The predominant x-ray-based methods are limited either by the high level of imaging dose used for fluoroscopic imaging of small implanted markers or by the snapshot nature of the imaging data, such as those provided by cone-beam CT (CBCT). [12][13][14] The tracking of implanted electromagnetic transponders (i.e., Calypso) avoids the use of ionizing radiation but is unsuitable for pancreatic cancer given the large size of the transponder and the invasive procedure needed to implant them. 15,16 The recently introduced onboard MRI radiation systems offer a powerful real-time, non-invasive, and non-ionizing solution to guide and monitor SBRT of soft-tissue targets such as pancreatic cancer. 17,18 However, it remains uncertain as to whether these advanced and expensive systems will be generally available to the community. As an alternative, ultrasound imaging has low cost, the ability for image enhancement with contrast agents, mobility to be shared among machines, and compatibility to add to any existing treatment room. [19][20][21][22][23] Ultrasound imaging has been previously developed for imageguided radiation therapy and is commercially available for prostate intrafraction monitoring. 24,25 However, with the exception of recent efforts from the active robotic arm, 26-31 ultrasound imaging-based intrafraction monitoring clinical studies are still limited to prostaterelated applications. This is mainly due to the lack of probe holders for other sites such as pancreas and liver and the lack of treatment planning method to accommodate probe placement during treatment. 27,28 In this article, we introduce an arm-bridge system for intrafraction real-time motion monitoring during pancreas SBRT. We validated the image guidance workflow with volunteer study and studied the ultrasound monitoring accuracy using an ultrasound phantom and motion stage. We also investigated the impact of the probe placement in the treatment planning. Figure 1 shows a block diagram of the workflow of our study design.

| METHODS
Our proposed arm-bridge system was validated through the image guidance workflow. It was scanned and then segmented from the CT images. Previously treated pancreas SBRT patient CT images were fused with the segmented arm-bridge system to create virtual simulation CT. Two types of treatment plans, both IMRT and VMAT, were generated by following our clinical pancreas SBRT protocol criteria and avoiding the probes in the virtual simulation CT. They were compared with the clinically treated pancreas SBRT with IMRT plans.
In addition, a phantom study and volunteer study were performed to evaluate the accuracy of US monitoring. Participation of human subjects in the study was approved by the Internal Review Boards (IRBs) of the Johns Hopkins University School of Medicine where retrospective plans were analyzed and healthy volunteers were recruited.

2.A | Arm-bridge system
Our goal is to monitor the pancreas motion during the SBRT. Therefore, one of our top concerns for designing such system is its interference with treatment delivery. The optimal system should have the minimal blockage for planned radiation beam delivery. However, the desired probe orientation should allow the maximum scanning volume rate from the ultrasound probe. To accommodate these requirements, we designed the probe holder system as an arm-bridge system.
The system consists of a couch top bridge, articulated arms, infrared tracker, and ultrasound probe case. The bridge has rails on the bottom, enabling it to be attached to different couch tops. Two passive arms are used in the design to allow both fast placement and fine-tuning of the probe position. Finally, a quick release mechanism on the probe case allows the user to detach the probe for freehand scanning. can then be fused to the planning CT in the Clarity image workstation. The short arm can be disconnected from the long arm from the adapter so that the user can operate the ultrasound probe freely during the initial scan. Once the optimal probe position and orientation is found by the user, the probe with the short arm can be connected back to the long arm. The user can then fine-tune the position and orientations of the probe (i.e., roll, pitch, and yaw) using the short arm.
The time taken to acquire an ultrasound image volume depends on the imaging depth (probe axial direction), the number of lines or sector width (probe lateral direction), and the mechanical sweeping angle or number of frames (probe elevational direction). Based on our clinical experience, the major motion for the pancreas is in the patient superior-inferior direction. In our design, the ultrasound probe is oriented so that the mechanical sweeping or the elevational direction is aligned with the patient left-right direction to minimize ultrasound acquisition time and allow maximized volume scanning rate in future studies. The ultrasound probe lateral direction is aligned with the patient superior-inferior direction in the treatment room.

2.B | Image guidance workflow validation
To validate our design in the setting of our pancreas patient image guidance workflow, we used the ABDFAN ultrasound phantom as F I G 1 . The study design workflow. The proposed arm-bridge system CT was validated for image guidance workflow. It was segmented and fused with previous patient pancreas CT to create the virtual simulation CT. The IMRT and VMAT plans from the virtual simulation CT were then compared with the prior clinically treated pancreas IMRT plan. To further validate the clinical setting, we tested the ultrasound monitoring system with volunteer study, in which three volunteers of different sizes were included. The volunteer and ultrasound system were set up on CT couch, and the couch was moved through CT bore to test clearance. Then, volunteer and ultrasound system were set up on treatment couch with ABC. The gantry was rotated to different angles to check clearance (Fig. 3).

2.C | Virtual simulation CT
To simulate the arm-bridge system during CT scan and the planning, we created virtual simulation CT by combining the armbridge system CT and previous patient simulation CT. We first setup the arm-bridge system together with an ultrasound phantom and then scanned both of them with CT. The ultrasound probe, probe case, infrared tracker, and the short arm were segmented from the CT image and then virtually placed on the patient

2.D | Virtual simulation treatment planning
The clinically treated pancreatic SBRT plans in our institute used 10 or 11 coplanar IMRT beams, and ABC was used during simulation and treatment to constrain target movement. The plans were deliv-   In volunteer study, the reproducibility of ultrasound monitored ABC was investigated. The volunteers were guided to do ten breathholds with real-time ultrasound monitoring. Superior mesenteric vein (SMV) was selected as monitoring ROI. The ultrasound imaging system performed real-time monitoring based on ROI during ten cycles of breath-holds. The positions of ROI for ten breath-holds were recorded and compared to get the reproducibility.

3.A | Image guidance workflow validation
The arm-bridge system and the ultrasound probe were validated in clinical image guidance workflow. Figure 5    The red lines are for PTV, green lines are for duodenum, blue lines are for stomach, dark red lines are for bowel. Three plans are F I G . 8. The 3D rendering of beam orientations for virtual simulation IMRT plan (left) and VMAT plan (right) of an example patient from this study in our planning system Pinnacle. The step-and-shoot IMRT plan consists of 10 beams, each beam has 5-10 segments. The VMAT plan is composed of two arcs; each arc has about 70 control points. Gantry angle interval between two consecutive control points is 2°. In both IMRT and VMAT plans, the beam is restricted as least 30°away from the probe.  For D95 and V33, the virtual simulation VMAT plan shows higher coverage than the virtual simulation IMRT plan. Based on OARs constraints (duodenum-V15/20, stomach-V15/20, liver-V12, and stomach-V12), as shown in Table 2, there is no major difference between the clinically treated plan and the virtual simulation plans.

3.C | Ultrasound monitoring accuracy and ABC
Our experiment proved good ultrasound monitoring accuracy of our system. Figure 11 shows the setup of ultrasound phantom, motion platform, and arm-bridge system in the simulation room (left) and the Clarity Guide real-time monitoring of the ultrasound phantom motion (right). After aligned to the laser with the premark on the phantom surface, a monitoring reference ultrasound image was acquired at the simulated exhale phase from the motion platform.
The monitoring module then started to monitor the 3D motion with time in left-right, anterior-posterior, and superior-inferior directions and real-time ultrasound image views. The phantom motion between inspiration and expiration captured by the camera (ground truth) is In the volunteer study, the reproducibility of 10 ABC breathholds of all three volunteers was less than 2 mm. Detailed results can be found in Table 7. The data indicate our system could potentially provide accurate tracking of soft tissue in clinical settings.

| DISCUSSION
While we are accumulating more clinical evidence to support the benefit of pancreatic cancer treated with SBRT, intrafraction treatment uncertainty due to motion may be potentially improved by using real-time ultrasound monitoring. In addition to the similar advantage of being non-invasive, non-ionizing, and real-time, T A B L E 3 Dosimetric parameters of all ten virtual simulated IMRT plans. D 95: the minimal dose to 95% of the PTV, D5: minimal dose to 5% of the PTV, Dmean: mean dose to the PTV, CI: conformal index, the ratio between PTV and volume receiving dose larger than prescription dose, HI: homogeneity index, the difference between D5 and D95, divided by Dmean. IMRT and VMAT radiation beams through the ultrasound sound probe, we found that the planning quality is not compromised. It is, therefore, possible to achieve the same planning quality as the clinical plans when the probe and the arm-bridge system are present.
However, there were several limitations to this study. Our design, mechanical clearance, imaging accessibility, probe stability, and deployment efficiency should be evaluated more comprehensively in a pilot study on patients. In addition, ultrasound imaging of the pancreas may not always be possible due to the bowel or stomach gas causing poor image quality. Patient education on diet at the initial consultation with nurses and physicians, and following diet restrictions are crucial to improving patient ultrasound imaging quality. During the phantom CT imaging in the simulation room and CBCT imaging in the treatment room, CT and CBCT images showed noticeable metal artifact from the probe as in Fig. 3. To mitigate the metal artifact from the probe, we have worked on strategies with promising results such as using a mock probe. 27 Other groups also developed CT and CBCT reconstruction algorithms to reduce general metal artifact as in recent studies. 32 A clinical implementation of metal artifact reduction algorithm from such studies can help us to improve the image quality result from CT and CBCT with an ultrasound probe in placement. The speed of sound correction is a known issue for the accuracy of ultrasound imaging. Several groups have studied the accuracy and potential impact on ultrasound imaging. 33 In this study, we have not discussed the potential image degradation from the exit dose from radiation beam. It would be interesting to determine the dose level that can potentially degrade the image quality or damage the probe. In our study with the phantom and simulation CT, we did not include the soft-tissue deformation introduced by the probe weight. Our future study with clinical patients will allow us to better understand the impact of the probe weight on soft-tissue deformation, treatment planning, and ultrasound image quality.

| CONCLUSION
Our ultrasound system can potentially be used for real-time monitoring during pancreas SBRT. The phantom study showed high monitoring accuracy of the system, and the volunteer study showed feasibility of the clinical workflow from high reproducibility of the ABC breath-hold. No planning quality compromise is required for pancreas SBRT treatment delivery with ultrasound imaging. Future studies will concentrate on the clinical trials with pancreas SBRT patients to optimize the clinical workflow for real-time ultrasound monitoring with our arm-bridge system.

ACKNOWLEDG MENTS
The authors thank Dr. Martin Lachine and David T. Cooper for their inputs in the Clarity software. This work was supported, in part, by grants from the National Institutes of Health (NCI R01 CA161613) and Elekta to J. W.

CONFLI CT OF INTEREST
The authors declare no conflict of interest.