In this talk, I will present two different works: one devoted to evaluating the aggressiveness of meningioma with a mathematical model and a second one devoted to having an evaluation of the risk of relapse of low grade gliomas using machine-learning techniques.

In the case of meningioma, several MR images are available for each patient, that allows us to build a spatial model based on a set of Partial Differential Equations (PDE) at the scale of medical images. This model is personalized for each patient and once its parameters are recovered from the two first exams, we let it run to obtain a prediction of the evolution of the tumor. The accuracy of the prediction is validated on a large cohort of patients selected by our collaborators at CHU Pellegrin, Bordeaux.

For low grade gliomas, only one time point, at diagnostic, before surgery, is available per patient. We cannot build an evolution model as in the previous work. Our goal is to correlate radiological features computed from this exam as well as clinical information with the speed of relapse. This yields an invaluable tool for clinicians to plan patients' followup. For this matter, we applied statistical learning techniques on classical radiological, biological and clinical features as well as new ones that we have specifically developed in this context. This algorithm is validated on more than 120 patients from Humanitas Research Hospital at Milan, Italy.