Brain tumours comprise a relatively rare disease, but one that is a significant cause of morbidity and mortality. Diagnosis of brain tumours, which is based mainly on magnetic resonance imaging (MRI), is crucial for choosing optimal treatment strategies. In many cases, however, MRI provides non-specific findings and differentiation of individual types of tumours may be problematic using conventional MRI methods. In recent years, various MRI diffusion imaging techniques have been introduced that, in conjunction with advanced data analysis, have the potential to upgrade the possibilities for non-invasive classification of brain tumours. The main goal of this project is to develop automated methodological procedures enabling the differentiation of individual types of brain tumours based on multimodal MRI data. The prospective study will include about 240 patients with brain tumours who will undergo MRI of the brain and subsequent resection or stereotactic biopsy of the pathological brain lesion. The MRI protocol will include diffusion tensor imaging (DTI) and intra-voxel incoherent motion (IVIM) techniques.

Using machine-learning methods, automated techniques of MRI image segmentation are being developed together with advanced classification algorithms enabling differentiation of individual tumour types based on their morphological and diffusion features. The accuracy of the stated procedures in the matter of differentiating individual types of tumours will be verified in correlation with histopathological findings. For training and validating segmentation methods, gold-standard reference annotations are being created by experts.

In collaboration with international partners, also federated learning approaches to brain tumour segmentation are being tested. In particular, we are part of the Federated Tumor Segmentation (FeTS) initiative led by University of Pennsylvania.