Session Item

Conventional radiation treatment planning strategies intend to maximise survival and quality of life (QoL) of cancer patients but are based on subjective interpretation of how QoL is related to the dose distribution. We aim to implement a closed loop between observed QoL and treatment plan optimisation and reduce the need of human interaction.

Prospective data of 750 patients treated for head and neck cancer were analysed to find NTCP models for 20 common toxicities at 6 to 24 months after treatment. Subsequently, a linear relation between those toxicities and QoL, defined as an average of six EORTC QLQ-C30 items, was fitted with use of dimensionality reduction (PCA) and resulted in a weight factor of each toxicity with respect to QoL. The NTCP models and their weights were then used in treatment plan optimisation. Models were programmed into the optimiser of RayStation using C++ and the Raysearch Research functions API with adaptations to make the functions convex and smooth to ensure convergence. Case specific settings of the objective functions were prepared in a spread sheet and imported in RayStation with a python script. Objectives for target coverage, hot spots, and conformity were manually adjusted iteratively until the plan was clinically acceptable (see figure). The planning study was conducted with 31 patients.

QoL-guided optimisation converged properly and planning required less human interaction than with conventional planning. QoL-guided plans showed improved predicted QoL with equal target coverage. The improvement was achieved largely by decreased sparing of the parotid glands and increased sparing of other normal tissues (see table). This can be explained because the parotid glands were historically the first organs to be actively spared and have retained a very high priority in conventional planning, while this priority seems unjustified in QoL-guided optimisation.