Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/377C8AB5-DB8B-4022-90FC-D711DAA126CB | ||
| Year | 2024 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Maria-Aikaterini Dourida, "Analysis of magnetoencephalographic signals from children with reading difficulties using realistic head modeling and machine learning.", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024 https://doi.org/10.26233/heallink.tuc.100268 | ||
| Appears in Collections |
Reading Difficulties are the most common type of learning disability, often manifesting at a young age, including deficits in reading comprehension, word decoding, orthography and phonology. Magnetoencephalography (MEG) is a neuroimaging technique that captures brain activity in millisecond precision. The source analysis of MEG recordings can provide an estimation of the underlying brain activity without being affected by volume conduction effects. An appropriate base to investigate functional connectivity patterns is thus established. In this thesis, resting state MEG recordings from 40 Non-Impaired (NI) children and 26 children with Reading Difficulties (RD) are analyzed with the goal to estimate the intra-, inter- and dominant frequency-based brain interactions and successfully classify them between NI and RD groups. Initially, source location is performed using beamforming techniques and realistic head modeling. Using intra-frequency and inter-frequency phase synchronization metrics on a time-varying fashion, source interactions are estimated. The dominant frequency coupling is also estimated to further reveal the maximum coupling source interaction. Symbolic time-series and their complexity index are then estimated. A comparison of the Inter-, Intra- and Dominant frequency approaches is conducted, by analysing the brain networks across brain regions, as well as their FCmstates. From the results, it is indicated that the dominant frequency (δ-β) provides a more accurate depiction of the differences between groups across brain regions and FCmstates. By employing the machine learning approaches k-Nearest Neighbors (κ-ΝΝ) and Support Vector Machine (SVM) on the dominant frequency, a high classification performance (Accuracy > 96 %) is observed. Highly accurate classification is also achieved across all the examined phase synchronization metrics. The present thesis thus paves the way for future non-invasive diagnostic systems for identifying reading difficulties in childhood.
Copyright © DIAS 2013
