ESTRO 2020 Abstract Book

S209 ESTRO 2020

SP-0391 TCP modelling for better radiotherapy plans I. Toma-Dasu 1 1 Karolinska Institutet and Stockholm University, Medical Radiation Physics, Stockholm, Sweden Abstract text Modern radiotherapy employs radiobiological models for calculating tumour control probability (TCP) and normal tissue complication probability (NTCP) at different points starting from treatment planning optimisation to treatment outcome evaluation. The modelling of the probability of controlling the tumour relies on the assumption that all the cells in the tumour need to be eradicated in order to prevent the local recurrence. Mathematically this assumption could be expressed as a function of the initial number of clonogenic cells in the tumour and the probability they will survive a certain radiation therapy regimen. There are, however, many challenges to overcome in order to adequately account for the different degrees of complexity of the tumour biology and physiology as well as for the complexity of the treatment in order to use a TCP model for a certain clinical application. This presentation aims to make a review of the TCP models accounting for the various levels of complexity and to illustrate their applications in modern radiotherapy. The modelling of TCP for a population of patients as well as for individual patients will be considered. Particular emphasis will be put on the latter in light of the increased interest in the clinical community for individualised radiotherapy going beyond dose conformity. One of the current trends in radiotherapy planning involves the use of heterogeneous target dose distributions. They might be the result of an optimised plan involving high modulation of the intensity of the beams, hence highly heterogeneous fluence distributions, aiming at increasing conformity of the prescribed high doses to targets of complex shape or in close vicinity of the organs at risk. The heterogeneous dose distributions might also be intentionally planned since the classic paradigm of homogeneous dose prescription to the target has been challenged and novel dose prescription approaches accounting for the spatial distribution of subvolumes consisting of cells with different sensitivities to radiation have been proposed. Thus, accounting for the heterogeneity of the dose distribution over the target is a prerequisite of any TCP model. The new trends on going beyond the heterogeneity in dose in characterising the energy deposition events leading to biological effects will be also discussed. In addition to the heterogeneity in the physical factors, one has to account in the modelling of TCP for the heterogeneity in the biological factors affecting the sensitivity to radiation and the overall tumour response, including the heterogeneity in the density of clonogenic cells and in the factors related to the complex tumour microenvironment. Tumours are dynamic systems with respect to their radiosensitivity and therefore modelling TCP should consider both the spatial and the temporal variations of the radiotherapy factors influencing the outcome. The applicability of TCP models for biological optimisation of the treatment plans is limited by the availability of their parameters. The overall accuracy as well as the sensitivity and specificity of the prediction of the outcome based on TCP models are also highly dependent on the model parameters. The parameters describing the dose-response in terms of probability of controlling the target could be empirically derived based on the response of a population

sophisticated multivariable NTCP models would be more clinically meaningful than the gEUD. Therefore, treatment plan optimization should be facilitated by multivariable NTCP models. The knowledge captured by NTCP models can be used to optimize treatment plans for novel patients. The flow chart in figure 1 illustrates a typical learning cycle in which NTCP models are trained and used for treatment plan optimization. A straightforward approach to optimize a dose distribution by means of NTCPs, is to directly translate multivariable NTCP models into an objective function, i.e., minimizing the objective value (i.e. the NTCP) directly by the optimizer. We have automatically optimized dose distributions for head and neck cancer patients with multivariable NTCP models for various clinical end-points, such as xerostomia and dysphagia. We demonstrated that multivariable NTCP-based objective functions combined with dose-volume objectives for the targets results in clinically realistic treatment plans. In addition, the NTCP-based objectives facilitates automated optimization, in which the search for a traditional dose threshold value was omitted and replace by the multivariable NTCP model directly. This approach is, however, not supported by the commonly used commercial treatment planning systems. An alternative approach is, therefore, to translate the dose-volume parameters from the NTCP models into traditional dose-volume objectives. The regression coefficients of the logistic regression coefficients of the multivariable NTCP models provide inside into the importance (i.e. the relative weight) of a particular OAR. Since 2007, the latter approach was followed in our clinic. The first years (2007-2011), 201 patients were treated with standard parotid sparing IMRT (ST-RT), while maintaining objectives and constraints for the target volumes and serial-like organs. The following years (2012-2017), 704 patients were treated with so-called swallowing sparing VMAT (SW-RT). Since 2017, 213 patients were treatment with so-called oral cavity sparing treatments using either VMAT or IMPT (OCS-RT). Following two cycles of the NTCP- guided optimization loop (figure 1), on average, the oral cavity D mean decreased significantly from 48.5 Gy (±15.7) with ST-RT to 27.7 Gy (±16.3) with OCS-RT (p<0.001). The D mean of the superior pharyngeal constrictor muscle significantly decreased from 57.6 Gy (±13.8) to 41.5 Gy (±18.9) from ST-RT to OCS-RT, respectively (p<0.001). Consequently, significant reduction in grade≥2 and grade≥3 physician-rated and patient-rated dysphagia as compared to PS-RT were reported 6 months after treatment.