Improving the Efficiency of On-line Dose Verification for Proton Lung SBRT

Objectives: : Proton beam has no exit dose. For lung SBRT, protons offer lower integral lung dose than Linacs, especially benefiting patients with prior radiation. However, protons are more sensitive to anatomy variations such as body surface displacement and lung density change in beam path. Therefore, it is essential to build an on-line dose calculation workflow to verify plan can hold up before delivering each fraction.

Our current clinical patient setup workflow for proton lung SBRT is presented below. First, therapists align the patient using a pair of orthogonal oblique X-ray images from ceiling-mounted X-ray panels. The robotic patient position system (PPS) performs the shift in 6 degrees of freedom (DOF) based on the 3D-2D registration that best matches bony anatomy. Next, PPS is rotated and extended to a fixed position for volumetric imaging by CT-on-rail (CToR). Therapists, a physicist and the attending physician collaborate to match CToR to planning CT and focus on the lung tumor CTV. In addition, they evaluate the anatomy variations that could affect the range of protons, such as body surface displacement, chest wall alignment, rib and scapula alignment, and lung density change. The care team estimate the dosimetric impact of those variation and verify that the plan is still robust to treat. Finally, patient is shifted back to post X-ray position and the shift derived from CToR volumetric match is applied.

Methods: : In our current automated off-line dose verification workflow, the CToR volume is resampled and re-gridded in the frame of reference of planning CT volume using the registration dicom file from R&V system, which accumulates the shift of X-ray 3D-2D match and CToR volumetric match. The re-gridded CToR is sent to TPS and the original plan is cast on that image for off-line dose calculation. However, this workflow is challenged in on-line dose calculation scenario due to two issues. First, registration dicom file is unavailable in on-line setting. It is not generated by image guidance system and sent to R&V until the care team proposes and applies the shift based-off CToR, which beats the purpose of using dose calculation to determine the best shift. Second, data transfer between systems (PACS, R&V, IGRT system) is time consuming, especially for CT and structure set dicom files, which constitute 30% of processing time in current off-line verification workflow.

To address these challenges, we proposed a new workflow engineered for on-line dose verification in this abstract. After CToR image acquisition, we immediately send CT dicom files to a Monte Carlo (MC) dose calculation engine, even though CToR does not represent the post-shift treatment position. Responding to the volumetric match proposed by the attending physician based off CToR, the physicist manually entered the proposed shift into a matlab program, which makes a copy of the original plan and modifies the delivery parameters to compensate for the shift. For instance, table translation is compensated by moving isocenter, table roll is compensated by changing gantry angle, in opposite directions. The modified plan is then sent to the MC engine for on-line dose calculation. Calculating the modified plan on the original CToR image is dosimetrically equivalent to calculating the original plan on the post-shift CToR volume. This approach is much more efficient as transferring a single plan dicom file is faster than transferring the re-gridded CT and structure set dicom files.

Results: : We built a dicom service to continuously run the matlab program for generating modified plans. To evaluate the efficiency of dose verification, we measure the time from the moment when CToR shift is proposed to when the calculated dose is ready for review. The current off-line workflow takes an average of 3.5 minutes — 2 minutes for R&V to receive incoming re-gridded image and structure set dicom files and perform its internal format integrity check and 1.5 minutes for analytic dose calculation. In contrast, the proposed on-line workflow takes approximately 1 minute — 10 seconds for plan modification and transfer, and 50 seconds on MC dose calculation. The efficiency is significantly enhanced by minimizing data transfer.

The algorithm of plan modification to account for the 6DOF patient setup is summarized as follows. Translational shifts are accounted for by changing the isocenter positions in the opposite direction, while the rotations (pitch, roll, yaw) are compensated by modifying gantry angle, collimator angle and couch angle. For each field, we calculate the transform matrix of CToR axes (relative to fixed coordinate system): M=R_z (couch angle)×R_z (yaw)×R_x (pitch)×R_y (roll), the transform matrix of beam-limiting device axes (relative to fixed system): B=R_y (gantry angle) and the transform matrix of beam-limiting device relative to CToR axes: M^(-1)×B.In order to calculate dose directly on the original CToR volume, the modified plan needs to maintain the same transform matrix of beam-limiting device relative to CToR axes. They are determined by analytically solving the follow equation:
(R_z (new couch angle))^(-1) × R_y(new gantry angle) × R_z(new collimator angle) = M^(-1)×B

Some example results are presented below. 3° roll only corresponds to -3° change of original gantry angle, while 3° yaw only corresponds to +3° change of original couch angles. For a lateral beam (eg. 90° gantry angle), 3° pitch corresponds to 3° collimator kick. For an AP beam at 0° gantry angle, 3° pitch corresponds to 90° collimator kick plus 3° gantry rotation. For any given combination of pitch, roll, yaw, couch angle and gantry angle, there exists at least one set of solutions, though the angles in the solution do not have a simple correspondence to the pitch/roll/yaw values because of the intricacies of the rotation matrix equation above.

Conclusion(s):: We developed a novel on-line dose verification workflow for proton SBRT. To provide on-line dose feedback to the proposed 3D-3D registration from CToR match, we modify the plan instead of re-grid CT image. With the time of data transfer minimized, this method significantly improved the efficiency of on-line dose calculation for proton SBRT.