1) Wider research context / theoretical framework Glioblastomas (GBM) are very aggressive brain tumours with poor overall survival and high local recurrence (LR) rates. Treatment concepts have remained almost unchanged for decades. The correlation of tumour imaging hallmarks with prognosis stratification and treatment response could allow personalized adapted therapies to improve outcomes. However, the GBM diagnosis, staging, and therapy choice based on MR has limitations. Positron emission tomography (PET) based on the amino-acid radiotracer O-(2)-18FFluoroethyl- L- Tyrosine (FET) can overcome MR drawbacks when differentiating LR from radiogenic alterations. The aim is to identify biologically active tumour tissues associated with LR in GBM to replace the homogeneous dose distribution conventionally delivered in RT by a dose distribution scaled based on the patient’s specific risk profile of LR. 2) Hypotheses/research questions /objectives This work aims to develop Artificial Intelligence (AI) models to predict LR from PET images and synthesize CT from MR images. Once the prediction models and synthesis networks are developed, they will be used to create an open-access, user-friendly software as part of the treatment planning system (TPS). 3) Approach/methods The rationale for predicting LR is the hypothesized correlation between the biological properties of a tumour imaged by PET or MR and the response to radiation therapy. For this purpose, the 200 GBM patients of the GLIAA trial and their already prospectively collected data (images and clinical output) will be utilized. Supervised AI models will be employed to predict LR, focusing on recurrence extension and localization. The generation of CT images using Generative adversarial networks (GAN) facilitates the integration of the results in RT workflow. CT images and resulting contours could be exported in commercially available TPS, minimizing multimodality registration errors, and CT could be avoided. 4) Level of originality/innovation Although the number of open-source software for automatic AI image segmentation is increasing, most of them are exclusively focused on CT or MR images. An open-source tool integrating multimodality automatic tumour segmentation and GBM prediction of LR, with an output image and structures compatible with commercially available TPS, constitutes a valuable tool to facilitate and optimize the RT workflow for GBM patients. Currently, it exists neither as a commercial nor an open-source tool. 5) Primary researchers involved The two project coordinators (doctoral students) will develop prediction models for LR and use GANs to synthesize CT from MR images. They will also contribute to the open-source software. Both the doctoral students will be supervised by Prof. Grosu, with his previous expertise in AI and medical image analysis.