A Convolutional Neural Network Framework for the Identification of Individual Neuronal Biomarkers based on Functional Brain Connectivity using Magnetoencephalography رسالة ماجستير

Abstract

Finding the neuronal biomarkers at the individual level is an overarching objective in neuroscience research. Neuroscientists were able to identify individuals based on their brain functional and structural connectivity as depicted from their magnetoencephalography (MEG), functional Magnetic Resonance Imaging (fMRI), or electroencephalogram (EEG) scans. This individual identification process is also referred to in research as individual brain fingerprinting. In this study, resting state MEG (rMEG) data of healthy individuals is provided by the Human Connectome Project (HCP). A novel approach is introduced towards individual brain fingerprinting by applying a deep similarity learning model, a so-called Siamese neural network including convolutional neural networks (CNNs), to functional brain connectivity (FC) metrics from rMEG. To prove the superiority of the deep learning approach, the performance is compared against a sophisticated machine learning algorithm, Support Vector Machine (SVM). Coherence and Amplitude Envelope Correlation (AEC) were used as the FC metrics for the SVM and the Siamese network. The Siamese network was able to outperform the machine learning model with an accuracy of 97% as compared to 81% accuracy coming from the SVM. In conclusion, convolutional neural networks are a very powerful computational tool that can boost analysis of any field, such as neuroscience. To this date, the research pool for FC fingerprinting has not yet benefited from the deep learning techniques, where this research can be used as the first step towards enhancing the analysis of FC fingerprinting.