Author: Ana Maria Barragan Montero, John A. Lee, Eliot Peeters, Romain Schyns, Edmond S. Sterpin, Sophie Wuyckens 👨🔬
Affiliation: UCLouvain, Universite Catholique de Louvain 🌍
Purpose:
Although the likelihood of a point being tumorous decreases with distance from the GTV, CTVs are still defined as binary masks. Recently, the concept of clinical target distribution (CTD), a probabilistic CTV, has been proposed as a better alternative. CTD assigns each voxel a probability of containing tumor cells, with this probability decreasing as the distance from the GTV increases.
This study aims to demonstrate, using the local dose deposition benefits of proton therapy, the potential for improving treatment outcome through a novel radiobiological optimization framework that combines CTD with TCP (Tumor Control Probability) and NTCP (Normal Tissue Complication Probability) models. The main challenge of this optimization framework is to handle the non-convexity of CTD based TCP/NTCP cost functions.
Methods:
A Gaussian function was used to expand the GTV by incremental margins of 1mm and such that after 5mm, the tumor probability is 5%. TCP and NTCP models were based on the linear surviving fraction with radiobiological parameters based on Buti et al. 2022. Probabilistic radiobiological objective functions and their derivatives were then calculated.
The optimizations on the TCP/NTCP cost functions was conducted in OpenTPS, with Scipy L-BFGS-B as the optimization technique, on a head-and-neck patient with 4 delivery angles in 30 fractions.
Results:
Compared to classical dose-constrained optimization techniques, with and without probabilistic target, the TCP/NTCP optimizations resulted in a similar reduction (up to 38%) of the Equivalent Uniform Dose (EUD) to OARs considered while maintaining adequate TCP. This reduction in dose is a direct result of incorporating NTCP considerations into the optimization process.
Conclusion:
This proof of concept shows that considering radiobiological endpoints directly into the optimization process along with probabilistic target volume definitions allows for new insights into the solution space, leading to treatment plans with lower complication probabilities.