ESTRO 2023 - Abstract Book

S1700

Digital Posters

ESTRO 2023

PO-1944 Investigation of a new type of objection functions for smoothing of LETdxDose

M. Schubert 1

1 Fraunhofer Institute of Industrial Mathematics, Optimization, Kaiserslautern, Germany

Purpose or Objective Currently, types of objective functions to include LETdxDose (LD) in proton therapy planning are limited. A reason is that, as the effect of LET is still a topic of ongoing research, accepted threshold values to prevent side effects are not as readily available as they are for dose. Nonetheless, there is the danger of high LD values in organs at risk (OAR) leading to adverse effects, even if they only occur in a small volume[1] . Thus, we investigated a new type of objective functions. Contrary to previously introduced objectives, it considers the spatial orientation of the voxels. The aim is to smooth the LD distribution and hence prevent hotspots. Materials and Methods We used convolution kernels for edge detection to obtain undesired structures of the LD distribution and consequently minimized them with a p-norm. The approach was tested with cases from the CORT [2] dataset. MatRad [3] was used to obtain the dose and LD influence matrices and optimization was done in python. To compare the effects, optimization was performed with or without objective functions to minimize or smooth LD. Results Table 1 shows the results for the OARs for the prostate case and the Laplace kernel. Using objective functions for smoothing, the total variance was reduced by more than factor 3. Moreover, as Figure 1 shows, smoothing led to a reduction of the highest values. For the rectum this effect was stronger than by minimization of LD.

Conclusion The introduced type of objective functions can smooth the LD distribution and consequently reduce hotspots. For future planning this enables the systematic avoidance of adverse effects. Due to the linearity of convolution, and the possibility