Abstract
Background and aims: In radiotherapy, accurate patient alignment before each treatment fraction is crucial to ensure the dose delivered to the patient corresponds to the physician-approved treatment plan. Currently, patients are aligned based on pre-treatment images employed as a surrogate for dose. In the era of synthetic imaging and fast GPU based Monte Carlo (MC) algorithms for proton dose calculation, we aim to examine whether automated and accurate target coverage can be obtained with dose-guided patient alignment on a clinical timescale.
Methods: Data of 20 oropharyngeal cancer patients treated with IMPT at our institute and with weekly verification CTs (vCTs)/ synthetic CTs corresponding to CBCTs collected during the treatment course was used. The relative position of vCTs to the planning CT (pCT) was optimized for target coverage using a gradient descent algorithm that recalculates doses with mocqui [1] on two NVIDIA A40 GPUs. It maximized the V95 of the main CTV of 70 Gy and the elective CTV of 54.25 Gy, both extended with 1 mm for optimization. For each vCT, 7 starting positions were chosen: 1 obtained from image registration to the pCT and 6 randomly chosen with a maximum distance of 15 mm in each of the directions to the image-registered position. The CTV V95 at the final position was compared to CTV V95 obtained with image registration in the TPS RayStation.
Results: The results are summarized in Fig. 1. The CTV V95 improved on average with 0.001±0.004 compared to image registration and the average optimization time was 161±50 s. The CTV clinical goal of V95 ≥ 0.98 was met after dose-guide optimization for all cases, including the 4 instances that did not reach the clinical goal with image-based alignment.
Conclusions: We demonstrated that an automated and accurate dose-guided patient alignment to CTV at clinical timescale is possible using GPU MC and 3D imaging. Further work is warranted towards clinical adoption for pre-treatment patient positioning.
[1] H. Lee, et al., moquimc, https://github.com/mghro/moquimc (2024).
Methods: Data of 20 oropharyngeal cancer patients treated with IMPT at our institute and with weekly verification CTs (vCTs)/ synthetic CTs corresponding to CBCTs collected during the treatment course was used. The relative position of vCTs to the planning CT (pCT) was optimized for target coverage using a gradient descent algorithm that recalculates doses with mocqui [1] on two NVIDIA A40 GPUs. It maximized the V95 of the main CTV of 70 Gy and the elective CTV of 54.25 Gy, both extended with 1 mm for optimization. For each vCT, 7 starting positions were chosen: 1 obtained from image registration to the pCT and 6 randomly chosen with a maximum distance of 15 mm in each of the directions to the image-registered position. The CTV V95 at the final position was compared to CTV V95 obtained with image registration in the TPS RayStation.
Results: The results are summarized in Fig. 1. The CTV V95 improved on average with 0.001±0.004 compared to image registration and the average optimization time was 161±50 s. The CTV clinical goal of V95 ≥ 0.98 was met after dose-guide optimization for all cases, including the 4 instances that did not reach the clinical goal with image-based alignment.
Conclusions: We demonstrated that an automated and accurate dose-guided patient alignment to CTV at clinical timescale is possible using GPU MC and 3D imaging. Further work is warranted towards clinical adoption for pre-treatment patient positioning.
[1] H. Lee, et al., moquimc, https://github.com/mghro/moquimc (2024).
| Original language | English |
|---|---|
| Publication status | Published - Jun-2024 |
| Event | 62nd Annual PTCOG Conference - Singapore, Singapore Duration: 10-Jun-2024 → 15-Jun-2024 |
Conference
| Conference | 62nd Annual PTCOG Conference |
|---|---|
| Abbreviated title | PTCOG62 |
| Country/Territory | Singapore |
| City | Singapore |
| Period | 10/06/2024 → 15/06/2024 |