ESTRO 38 Abstract book

S1093 ESTRO 38

dose to OARs while ensuring robust target coverage. This will most likely translate to a clinical benefit for the patient. EP-1998 Quality assurance criteria and anatomically plausible models for deformable image registration C. Zachiu 1 , B. Denis de Senneville 2 , B. Raaymakers 1 1 University Medical Center Utrecht, Department of Radiotherapy, Utrecht, The Netherlands ; 2 Université de Bordeaux, Institut de Mathématiques de Bordeaux, Bordeaux, France Purpose or Objective Concepts in adaptive radiotherapy such as contour propagation and dose accumulation often rely on deformations provided by medical image registration algorithms. However, existing registration methods are often built upon elements originating from the computer graphics domain. Therefore, the anatomical plausibility of the estimated deformations may not be guaranteed. In this work we propose two quality assurance (QA) criteria, which assist in evaluating the anatomical plausibility of the deformations estimated by registration algorithms. In addition, a novel deformable image registration method is also proposed, specifically designed for biological soft tissue tracking. Material and Methods We propose assessing anatomical plausibility via two criteria: the jacobian determinant and the curl magnitude of the estimated deformations. Due to their high water content, most biological soft tissues are incompressible and therefore the jacobian determinant of their deformations is close to one. On the other hand, the local curl magnitude of the deformations should be close to zero, since local torsions inside most organs are unlikely to occur. Two methods are evaluated against the proposed criteria: the state-of-the-art EVolution registration algorithm (EVO) and a novel method “EVolution Incompressible” (EVI) proposed here. The latter replaces the smoothness regularization of the former with a penalty on deviations from unity of the jacobian determinant of the deformations. By doing so, the new method constrains the estimated deformations to be incompressible. The two methods were employed for the registration of five abdominal data batches, with each batch including five pairs of: (1) T1w MR – T1w MR, (2) CT – CT, (3) T1w MR – T2w MR, (4) CT – CBCT, (5) CT – MR images. The deformations estimated for the liver and kidneys were assessed both in terms the proposed QA criteria and the dice similarity coefficient (DSC). Results Table 1 reports the average DSC for the liver and kidneys before and after registration, with both methods leading to nearly identical improvements. However, a statistical analysis of the jacobian determinant of the deformations (see Fig. 1(a)) reveals high deviations from unity in the case of the EVO algorithm. A Mann-Whitney test indicates that this effect is significantly dampened by the EVI method at p = 0.05 . Fig 1(b) illustrates an analysis of the vorticity of the deformations estimated by the two methods, with EVI providing overall lower values for the curl magnitude.

Conclusion Both methods demonstrated a comparable performance for contour alignment. However, within the organ boundaries, the proposed method showcased an improved anatomical plausibility, with significantly lower compressions/expansions and vortices. This demonstrates both the importance of the proposed QA criteria for assessing the anatomical plausibility of deformations within the organ boundaries and that of adapting the registration model to the material properties of the observed tissues. EP-1999 Robustness of IMPT plans towards anatomical variations for nasopharyngeal cancer R. Argota Perez 1 , M.B. Sharma 1 , K. Jensen 2 , A. Vestergaard 2 , J.B.B. Petersen 3 , S. Korreman 3 1 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark ; 2 Aarhus University Hospital, Danish Center for Particle Therapy, Aarhus, Denmark ; 3 Aarhus University Hospital, Danish Center for Particle Therapy and Department of Oncology, Aarhus, Denmark Purpose or Objective Proton dose delivery is very sensitive to any variations during the treatment course. Setup uncertainties can be accounted for using robust optimization, whereas anatomical variations, e.g., changes in weight, tumour shrinkage, and filling or emptying of nasal cavities for head and neck cancer patients, are usually not accounted for. The present study evaluates the performance of a standard robust optimization technique towards Proton treatment plans using simultaneous integrated boost (68/60/50 Gy) were optimized for five patients with nasopharyngeal cancer using the Eclipse TPS vs 13.7. Planning was performed using multiple-field optimization (MFO) with three beam directions, and robust optimization for the 68 Gy CTV (CTV1) with 5 mm perturbations in all directions. Anatomical changes were simulated by density overrides in the original CT scans. For weight gain and loss, 5 mm of water equivalent material (WEM) was added or subtracted to the body contour. Nasal cavity filling was simulated by overriding the entire cavity density with WEM. Setup errors were simulated using isocenter shifts of +/-5 mm in three orthogonal directions, separately and in combination with anatomical changes. The dose was recalculated with the original plans in each simulated situation, and plan robustness was evaluated comparing CTV1 coverage (V 95 ) and OAR doses. anatomical variations. Material and Methods