Through the development of this training-through-research project, the fellow expects to acquire new skills, knowledge and perspectives, as well as to develop and widen her competences significantly, all contributing to her career development. The purpose of this proposal is to acquire new research expertise in Bayesian methodologies applied to source identification in blind signal separation and applied also to fusion of different modalities of physiological measurements for tumour delineation in brain.Magnetic resonance (MR) is key for the non-invasive analysis of brain tumours in the field of neuro-oncology. MR imaging (MRI) provides a morphologic characterisation of tissues, while MR spectroscopy (MRS) provides their biochemical information, resulting in precise metabolomic signatures. In this project, a multimodal MRI and MRS data analysis using Bayesian methods is proposed to address some medical questions that remain open, such as using additional knowledge to help extract better sources from the MR spectra, identify the exact number of underlying tissue types present in a sample and their spectroscopic patterns, and the use of them to track response to therapy. To address them, it will be necessary to solve some challenges from the methodological viewpoint, such as the extraction of relevant source signals in a multimodal (MRI and MRS) unsupervised approach, and the identification of the most appropriate number of source signals. A novel Bayesian approach of NMF tailored to facilitate the unsupervised multimodal analysis of MR data is planned to be developed, since Bayesian models deal with multimodal systems in a natural way, and the relationship between variables is explicit, as well as the handling of the prior knowledge. Data from pre-clinical models and human data from the European project “ETUMOUR” will be used to test the models developed as part of this project.