ESTRO 38 Abstract book

S546 ESTRO 38

2 Willis-Knighton Cancer Center, Radiation Oncology, Shreveport, USA ; 3 Ion Beam Applications, Research and Development, Louvain-la-Neuve, Belgium Purpose or Objective Scanned proton therapy is a highly conformal treatment technique and can provide increased healthy tissue sparing compared to photons or scattered protons. However, due to the time structure of the scanned beams this technique is highly sensitive to uncertainties and needs to be monitored when treating mobile tumors. Currently, only 3D Cone Beam CT (CBCT) acquisition is available on commercially available proton gantry systems. We investigate the possibility to use the same number of projections as for 3DCBCT acquisition to retrospectively reconstruct into 4D, with the goal to determine the inter-fraction tumor motion. Material and Methods CBCT-projection data was acquired for eight patients at an IBA Proteus® ONE proton gantry system. Ten 4DCBCT images were reconstructed per patient. Iin order to compensate for the lower 4DCBCT quality due to the small amount of projections per phase, we applied a motion- aware temporal and spatial regularization method (MA- ROOSTER), by Mory et al 1 , to reconstruct with improved 4D-image quality. Corresponding 3DCBCT images were reconstructed using the conventional Feldkamp-Davis- Kress (FDK) algorithm. After importing the 4D-images into RayStation, the gross tumor volumes were deformably warped (ANACONDA algorithm) and centroid positions determined to calculate the 3D-vector motion[OLd(1] . The contrast-to-noise ratio between tumor and lung tissue for the 4DCBCT images was calculated. Additionally, the structural similarity index (SSIM) between the 3DCBCT and 4DCBCT images were calculated to compare the quality of 3DCBCT vs. 4DCBCT images. Results Figure 1 shows one phase of the reconstructed 4DCBCT images for three patients compared to 3DCBCTs. For the 4DCBCT images more noise is observed, and increased blurriness of anatomical features with more clearly present streak artefacts for some of the patients. Nevertheless, the tumor is visible and can be delineated to evaluate its motion. Table 1 shows the calculated 3D- vector centroid motion, the CNR, and SSIM for the 4DCBCT images. In all 8 patients motion variations could be tracked within all available ten scans. The CNR showed varying results for the different patients, confirming the observed variations in image quality when visually evaluating the reconstructed 4DCBCTs. The quality of the 4D-images according to the SSIM was around 30% of the 3DCBCT images (SSIM ~0.30).

Conclusion It is possible to reconstruct 4DCBCT images with sufficient quality for tumor motion evaluation using the MA- ROOSTER algorithm. This is an important step towards an image guided adaptive proton therapy workflow for treating moving targets, however further work is warranted to improve 4DCBCT image quality. PO-0993 Uncertainty estimation of dose accumulation with deformable image registration in head and neck region T. Kanehira 1 , S. Kranen 1 , J. Sonke 1 1 The Netherlands Cancer Institute, Radiation Oncology, Amsterdam, The Netherlands Purpose or Objective The delivered dose distribution typically deviates from planned dose due to anatomical changes. Dose accumulation estimates the delivered dose over fractions using deformable image registration (DIR) to take anatomical changes into account. Registration errors, however, introduce an uncertainty in the dose mapping. This study estimates the uncertainty in the accumulated dose and compares it with differences between planned and accumulated dose for head & neck cancer (HNC) patients. Material and Methods In this study, five HNC patients were included with a prescribed dose of 70 Gy in 35 fractions. Patients were positioned with daily CBCT guidance and received a repeat CT (rCT) mid-treatment for adaptive re-planning. CBCT- to-CT DIR was performed using an in-house implementation of bSpline deformations. To estimate the spatial uncertainty of the DIR, a modified version of the distance concordance metric (DDM) was implemented [1]. To that end, the deformable vector fields (DVF) deforming the plan CT (pCT) to the daily CBCT and the daily CBCT to the rCT were concatenated. Subsequently, the DDM for each voxel of the pCT was calculated as the standard deviation (SD) in all three directions over the typically 34 DVFs (we did not use the daily CBCT which was acquired on the same day as rCT) divided by the square root of 2. DDM was restricted to voxels within the patient external. To estimate accumulated dose, assuming the DIR uncertainty followed Gaussian distribution which had a SD of the DDM, we calculated the average dose over 35 samples from a 3D Gaussian distribution around each voxel. This was repeated 1000 times to calculate a confident interval of the accumulated dose. The dose uncertainty was calculated as a 95% confident interval (dose-CI) and any difference exceeding this dose-CI was considered as a true anatomical induced dose difference. For each patient, we calculated the accumulated dose, difference with planned dose (ΔD, accumulated-planned), dose-CI, and the percentage of voxels with ΔD exceeding the dose-CI. Results DDM was larger in soft tissues and close to the field-of- view borders (Figure 1): median of 0.07 cm and 95 th percentile of 0.37 cm within the external contour. These