ESTRO 2020 Abstract book

S331 ESTRO 2020

Poster discussion: PH: Adaptive radiotherapy and inter- fraction motion management 2

PD-0552 Validation of an average anatomy model based on deformable image registration for lung cancer G. Lim 1 , S. Van Kranen 1 , J. Sonke 1 , P. Remeijer 1 1 Netherlands Cancer Institute, Radiotherapy, Amsterdam, The Netherlands Purpose or Objective Image-guided adaptive radiotherapy based on cone beam CT (CBCT) is used to detect differences with respect to the planning CT (pCT). Significant deformations are typically mitigated by creating a new treatment plan based on a repeat CT (rCT). The average anatomy model (AAM) is an adaptive strategy in which CBCTs from previous fractions are deformably registered to the pCT to estimate systematic deformations [1,2]. By averaging the deformable vector fields (DVFs) a new synthetic CT (sCT) can be created, in which the systematic deformations are mitigated. We aim to apply the AAM to lung cancer patients in which a systematic offset between initial tumor and lymph node positions can occur, typically several mm’s [2]. Here we present a validation study of the deformable image registration (DIR) and AAM algorithms, required for clinical introduction. Material and Methods Two validation methods were used to assess the geometric reproducibility of the DIR and AAM algorithms. • A modified version of the distance discordance metric [3] was evaluated on a set of 10 patients, each with 5 CBCTs and an rCT. Each CBCT was deformably registered to the pCT as well as to the rCT, producing two DVFs that were concatenated to derive a corresponding pCT-rCT DVF. Ideally, the resulting 5 pCT-rCT DVFs per patient are all identical, and differences between them are a measure of the registration error of the DIR algorithm. This error was quantified by the sample standard deviation (SD) per voxel of all 5 pCT-rCT DVFs per patient.

•

The AAM was used to generate an sCT for 16 patients based on 4-5 CBCTs per patient. The AAM was then reapplied a second time but using registrations to the sCT instead of the pCT, resulting in a second sCT with a corresponding average DVF (DVF 2 ). Ideally, both sCTs are equal since they are derived from the same set of CBCTs, and any residual values in DVF 2 are a measure of the registration error of the full AAM algorithm.

Figure 2: Transverse profiles of single (left), multiple (center) and modulated (right) focused proton minibeams at entrance (top) and 98 mm depth (bottom). Conclusion These results suggest that magnetic focusing is a viable method of proton minibeam production. In this case a permanent Halbach quadrupole magnet assembly was used to generate the minibeams distribution without the need for cooling or complex power supply/control systems. Multiple minibeams could be delivered though precise motion control of the robotic patient positioner, while beam modulation can be achieved via control of either the incident proton energy or through the use of a range modulator.

In both methods, only voxels in the intersection of the convex hull around the lungs and all CBCT field-of-views were considered. Results The probability distributions of the SDs per voxel of the three vector components (RL: right-left, CC: cranial- caudal, PA: posterior-anterior) and vector length (VL) of all pCT-rCT DVFs per patient are shown in Figure 1. Similarly, the voxel displacements in DVF 2 are shown in Figure 2.