Heuristic analysis of multimodal data and its application to nuclear medicine neuroimaging

Andrea Chincarini
Avatar, no personal image available
Andrea Chincarini

Research area

Applied physics


This works tries to find the general link between two set of data acquired with different modalities and sharing a common identity label. The link is heuristically determined by training a Deep Neural Network (DNN), whose architecture is optimized on the whole dataset using genetic programming. We shall also define the statistical criteria for determining the significance of the link. We apply the methodology to a set of Positron Emission Tomography (PET) brain images and a set of Electro-EncephaloGraphic (EEG) records, acquired on the same cohort of subjects. The intent is to establish the presence of a link between PET & EEG; more specifically, whether the EEG data can determine the PET information and to which extent. The answer to this question is of paramount importance in the case of the amyloid-PET, that is a PET scan acquired with a tracer that binds the amyloid protein in the brain. Should we find a significant link to the EEG, it would have a dramatic implication in the early diagnosis of neuro-degenerative diseases.