Adaptive Radiotherapy: Deep Learning for the Analysis of Daily Imaging

In image-guided radiotherapy, the focus has consistently been on patient-specific planned radiation and accurate dose deposition in the patient. In adaptive radiotherapy, the focus is placed on the patient's interfractional, anatomical changes in order to enable the daily optimized dose application. In most cases, conventional radiotherapy is fractionated over several weeks, so that weight loss, for example, can have a significant influence on the optimized dose distribution. Daily cone-beam CTs (CBCT) are used for patient positioning and provide data on the daily anatomy and, consequently, the dose distribution in the patient. In the case of interfractional changes, a distinction can be made between anatomy that undergoes daily changes due to fluctuations in bladder or bowel filling levels, for example, and a continuous change such as weight loss or a decrease in tumor volume. 

The aim of the project is to develop and validate neural networks of artificial intelligence in order to enable an independent and rapid assessment of whether a recalculation and re-optimization of the dose distribution is necessary due to anatomical changes. This should improve safety in the radiation planning process and potentially lead to a reduction in the safety margin surrounding the target volume to be irradiated, which could subsequently reduce the likelihood of side effects. The results of the project should provide medical staff with a well-founded decision-making aid for daily radiation planning to ensure that the patient-specific dose distribution is correctly applied to the patient's current anatomy.