Margriet Kwint

Chapter 8 158 to verify and adjust the current 3DCRT-based NTCP-models for these new treatment techniques. A method to validate current NTCP-models is the use of real-world data. With the introduction of an electronic toxicity scoring registration (based on the CTCAE), a more systematic recording of treatment related toxicity was implemented at the NKI in 2012 (53, 54). In chapter 7, data from the electronic registration of AET was used to validate the applicability of real-world data for the NTCP-models for AET of CCRT for NSCLC-patients, for the 2 sequential cohorts as described in chapter 2 . We found that real-world data is a useful method to audit NTCP-models in clinical practice. This model should ideally be tested in several institutions. To further optimize NTCP models, patient reported outcomes (PROs) can be used, since PROs have shown to be an useful complement to improve precision and accuracy when comparing these to clinician reported outcomes alone (55, 56). To conclude, due to continuous improvement of radiotherapy techniques and schedules, a constant update of NTCP- models is crucial to adequately predict radiotherapy induced toxicities. The use of real-world data and PROs simplifies and accelerates quality assurance and provide tools to validate these NTCP-models. Future Perspectives Adaptive radiotherapy and artificial intelligence The studies in chapter 4 and 5 showed that tumor regression is frequently seen during treatment on CBCT. As already described in a previous paragraph, adaptive radiotherapy is a promising approach to account for tumor volume regression and other intra thoracic anatomical changes during treatment. Currently, adaptive radiotherapy is a time-consuming strategy due to the manual clinical effort needed to produce an adaptive plan in a short time period. Besides, in current clinical practice, the decision to make an adapted plan usually relies on the decision of the radiation oncologist and physicist, which is inherently subjective (57). Adaptive radiotherapy may benefit from the advances of artificial intelligence (AI). With the use of AI, sophisticated predictive models can be developed with observational data by a computer. Then, these models can assess complex relationships between data. Automatic segmentation based on deep learning is an example of a frequently used application developed by AI in radiotherapy treatment planning with the use of the RT-planning CT. With automatic segmentation, the delineation of OAR is accelerated since it is done automatically instead of manually. Besides, it decreases inter- and