Institutional Repository [SANDBOX]
Technical University of Crete
EN  |  EL

Search

Browse

My Space

Magnetoencephalogram analysis of subjects with mild head injuries using multilevel connectivity networks and graph neural networks

Kavvouras Sotirios

Full record


URI: http://purl.tuc.gr/dl/dias/8E5AE3E3-034D-4254-8977-B828592FBC45
Year 2024
Type of Item Diploma Work
License
Details
Bibliographic Citation Sotirios Kavvouras, "Magnetoencephalogram analysis of subjects with mild head injuries using multilevel connectivity networks and graph neural networks", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024 https://doi.org/10.26233/heallink.tuc.99138
Appears in Collections

Summary

Mild Traumatic Brain Injury (mTBI) is a common neurological condition with significant cognitive and functional implications. This thesis presents a comprehensive investigation into the neural alterations associated with mTBI by employing advanced neuroimaging techniques and machine learning. Specifically, we employ Multi-Layer Functional Connectivity (MLFC) analysis, Machine Learning (ML), and Graph Neural Networks (GNNs) to unravel the intricate patterns of brain network disruption and provide insights into the underlying neurophysiological mechanisms. Magnetoencephalographic (MEG) recordings were acquired from mTBI patients and healthy controls during resting states. MLFC captures multi-layer correlations in different frequency bands, revealing subtle connectivity changes between brain regions. ML classification demonstrates the potential to discern mTBI patients from controls based on neural features. GNNs model brain regions as a graph, capturing complex interactions and non-linear relationships. Integrating GNNs enhances our understanding of mTBI-related disruptions, providing a more holistic perspective. Our study enhances insights into altered functional connectivity in mTBI patients. Although GNNs exhibit significantly superior performance compared to traditional machine learning methods, achieving an accuracy of approximately 97% versus 80-85%, the application of MLFC presents less definitive outcomes, with results appearing notably ambiguous, ranging between 50% and 65%. The fusion of MLFC, ML, and GNNs unveils nuanced dynamics not captured by traditional methods. These findings contribute to understanding mTBI pathophysiology and may guide personalized interventions.

Available Files

Services

Statistics