Parkinson’s disease (PD) is a progressive disorder of the nervous system characterized by the degeneration of dopaminergic neurons found in the substantia nigra. In general, the diagnosis occurs once the patients start experiencing the well-known motor symptoms of this disease, namely stiffness, akinesia and resting tremor. By this moment, it is estimated that 60 to 80% of the dopamine-producing neurons have already been lost or impaired.However, as the loss of these neurons disrupts the functioning of the subcortical structures, many non-motor symptoms such as visual and olfactory disturbances, mood disorders, or changes in cerebral perfusion occur earlier in the pathology.In this context, our project aims to find specific signatures in the newly diagnosed Parkinson’s patients that can lead to earlier diagnosis and the classification of patients into sub-types for more tailored treatments to slow down the disease process.To reach this objective, we have explored three different approaches. Firstly, we searched for morphometric changes in structural data using popular techniques such as VBM, DBM and SBM. Secondly, we conceived a deep-learning methodology to detect brain diffusion anomalies employing reconstruction errors of a trained auto-encoder. Finally, we developed an original statistical approach based on mixture models generated from Student distribution laws to construct a reference control model from multi-modal quantitative data of “normality” and to classify the abnormalities present in our patients.