Margriet Kwint

General discussion and future perspectives 159 8 intra-observer variability (58). The challenge is to make this automatic segmentation suitable for CBCT’s, so that a new RT-planning CT is not mandatory to adapt radiotherapy treatment planning. Moreover, the use of AI in automated treatment planning is rapidly developing (59), which enables the opportunity to implement adaptive radiotherapy in clinical practice in the broader radiotherapy community, since for some departments the time consuming (manually) procedure of re-planning is currently a limiting factor for routine clinical implementation. The introduction of magnetic resonance-guided radiation therapy (MRgRT) enables the use of imaging with superior soft tissue contrast without the use of ionizing radiation. This allows real-time imaging of the tumor and OAR before and during treatment, and combined with an online-adaptive workflow, online adaptation of the RT-planning is possible (60-62). MR-guided online adaptive radiotherapy (MRgART) improves visualization for image guided and adaptive radiotherapy compared to the image quality of CBCT. Further research to develop AI techniques in IGRT and MRgRT should be embraced to improve efficiency of adaptive radiotherapy in daily clinic. Prediction modelling and radiomics TNM-staging is the cornerstone in the classification of lung cancer, based on comprehensive evidence from (randomized) clinical studies and observational data (63). However, there is still a considerable variation in the treatment response among patients with identical lung cancer stages. The search for new biomarkers to improve current prediction models for a better patient selection is therefore essential. In recent years, knowledge on tumor heterogeneity is increasing (64), both between and within tumors, urging the need for further developing treatment options that can operate on an individual level rather than on a population level. One of the new promising fields that uses this tumor heterogeneity in prediction modelling, is radiomics. Radiomics is a method that extracts large amounts of features from radiographic medical images using data-characterization algorithms. These features are subsequently correlated with tumor characteristics and/or prognostic endpoints using advanced machine learning algorithms to develop computational prediction models (65, 66). All lung cancer patients undergo medical imaging, especially CT- imaging, subsequently making CT-based biomarkers attractive to optimize current prediction models for NSCLC-patients. Fornacon et al. (67) found 43 CT-image based articles in which the prognostic or predictive role of radiomics signatures in NSCLC-patients were described. The conclusion of this review was, that generally the