ESTRO 2020 Abstract book

S335 ESTRO 2020

subjected to stringent quality assurance (QA) measures in order to ensure patient safety and therapeutic efficiency. In the present study we propose a set of novel QA criteria for DIR algorithms, based on biomechanical tissue properties. Material and Methods The proposed QA criteria imply the voxel-wise calculation of the mechanical normal and shear stress that would occur within the observed tissues as a result of the estimated deformations. These stress values are compared to a threshold (determined by previous studies via mechanical tests), which if exceeded would lead to tissue rupture. Similarly, compressions which exceed the typical arterial blood pressure are also detected. Since neither the “rupture stress threshold” nor the “blood vessel occlusion threshold” are expected to occur during typical anatomical motion at rest, it is highly likely that such voxels have been misregistered. Three DIR algorithms were employed for registering a pair 3D MR images of hepatocellular carcinoma, undergoing a set of known deformations generated via finite element modeling (FEM). The evaluated algorithms consisted of the open-source Elastix registration software and two in-house developed methods (referred here as EVO and EVI). Aside testing the estimated deformations against the proposed criteria, we have also investigated the contour propagation and dose warping accuracy of the algorithms. Results While all three registration solutions have demonstrated similar contour propagation capabilities, the estimated deformations have violated to different extent the proposed biomechanical criteria (see Fig. 1). Furthermore, when used to warp a simulated radiation dose of 20 Gy to the tumor, the algorithms showcased various amounts of differences relative to the dose warped via the known FEM-generated deformations (see Fig. 2). Of note here is that there is a spatial correlation between the areas “flagged” by the proposed QA criteria and errors within the warped dose maps.

Results When using the mean shape calculated from five scans, the variability in the remaining dataset decreased on average by 28% per patient for bladder, 20% for rectum, and 27% for prostate. For most patients, most of the organ motion was captured if averaging over four scans, with no significant variability reductions with additional scans (Fig. 2a). For many patients, the variability gradually decreased with number of scans used for mean shape computation. However, for some patients only minimal, or no stable decrease could be observed. For example, in the case of bladder shape variability we observed four patient groups: low stable variability reduction (<30%) in eleven patients, medium stable reduction (<40%) in nine patients, high reduction (>40%) in two patients, and two patients with no stable variability reduction (Fig. 2b).

Conclusion Using multiple pre-treatment CBCT scans captured over one week (four or five consecutive scans) as a basis for treatment planning can account for a significant portion of organ shape variability occurring during treatment. PD-0557 Biomechanical quality assurance criteria for deformable registration in image guided radiotherapy C. Zachiu 1 , B. Denis de Senneville 1 , M. Ries 2 , B.W. Raaymakers 1 1 University Medical Center Utrecht, Department of Radiotherapy, Utrecht, The Netherlands ; 2 University Medical Center Utrecht, Imaging Division, Utrecht, The Netherlands Purpose or Objective During image-guided radiotherapy (IGRT), deformable image registration (DIR) has the potential to automatically estimate and compensate for anatomical deformations by the means of both contour propagation and dose accumulation. Since future adaptive therapy strategies might include this information in the clinical decision- making process, the estimated deformations must be