Learning strategies for MRI Surrogates in Brain Imaging and BCI

The purpose of this thesis is to develop an algorithm, which based on some selected parameters, is able to extract from a worldwide database of MRIs, the magnetic resonance closest to the satisfaction of these criteria in other words it can provide a 3D model of a patients’ brain without the specific subject MRI. The first chapter consists of a brief introduction to the anatomy of the brain and the mechanisms of impulse transmission in the network of neurons, then the main neuroimaging techniques are explained, with particular attention to the magnetic resonance that is intended to be avoided with the developed algorithm. Although this method is very useful, it has numerous limitations, over all waiting times and operation costs. The chapter also describes the electroencephalography (EEG) which, when suitably analysed by solving two mathematical problems (forward and inverse problem), makes it possible to test the effectiveness of the algorithm. Finally the BCI Technology, also known like Mind Machine Interface (MMI), is mentioned, which is a technology that could give practical application to the algorithm itself. Within the second chapter the main selection parameters among the various MRIs are examined, particular attention has been paid to age, volume, and sex. The database used for training and testing is an Open Neuro sub-archive called LONDON HEALTH SCIENCES CENTER PARKINSON'S DISEASE DATASET, containing 40 functional MRIs, already divided by sex and age. A part of the chapter is dedicated to explain how the volume is obtained by interpolation between external and internal dimensions and how to get the volume starting from an MRI using the SIENAX software. In order to understand whether the parameters above selected were relevant, and which of those was the most significant for the algorithm, an analysis of the variance was carried out. The algorithm code is also introduced and explained in the chapter. In the third chapter we try to verify the validity of the code by solving the inverse and forward EEG problems in order to find a similarity between the patient and the selected MRI in terms of sources and electrodes activation. The matrices regarding the mathematical problems were obtained from the Brainstorming software. Brainstorm is an open-source, collaborative application dedicated to visualize and processing of EEG data, with particular emphasis on estimation techniques of cortical sources. The inverse problem has no unique solution because the data recorded by the EEG are not sufficient to locate the sources, an additional criterion for choosing one of the inverse solutions used in the project is sLoreta suitably treated and described in the chapter. Finally, the fourth chapter analyses the results obtained, leaving some insights to possible future developments focusing particularly on the morphological check.