Graduation Semester and Year
Summer 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Biomedical Engineering
Department
Bioengineering
First Advisor
Dr. Elizabeth M. Davenport
Second Advisor
Dr. Sasha Alick-Lindstrom
Third Advisor
Dr. Joseph A. Maldjian
Fourth Advisor
Dr. Georgios Alexandrakis
Fifth Advisor
Dr. Chandan Ganesh Bangalore Yogananda
Abstract
Epilepsy and dementia are debilitating neurological disorders that pose substantial challenges for patients, caregivers, and healthcare systems. Advances in magnetoencephalography (MEG) and signal processing offer new opportunities to improve diagnostic accuracy, surgical planning, and treatment monitoring. This dissertation presents a unified body of work comprising artifact removal, automated event detection, and deep learning-based biomarker discovery. These approaches collectively enhance the clinical utility of MEG for diverse patient populations.
The first study addresses a significant technical obstacle in the management of drug-resistant epilepsy. Patients receiving responsive neurostimulation (RNS) have historically been excluded from MEG as the data is contaminated by device-related artifacts, which complicates accurate localization of the seizure onset zone. This work introduces an automated independent component analysis (ICA)-based pipeline implemented in MNE-Python that is combined with the traditional temporal signal-space separation. Compared to conventional preprocessing, this ICA-based approach yields substantial improvements in signal-to-noise ratio and supports more precise dipole localizations that align with invasive monitoring and surgical targets.
The second study proposes an automated image-based deep learning pipeline for detecting interictal spikes in clinical MEG recordings from epilepsy patients. Sensor-space signals and topographic heatmaps were encoded as RGB composite images and classified with a pre-trained ResNet-18 convolutional neural network. The developed model was evaluated on an independent held-out test set, achieving an accuracy of 85.1% and an area under the receiver operating characteristic curve (AU-ROC) of 0.93. This streamlined architecture reduces the burden of manual annotation, improves reproducibility, and supports more efficient pre-surgical workflows.
The third study explores MEG as a functional biomarker for the early detection of dementia. Resting-state MEG data from healthy controls, mild cognitive impairment, and dementia were transformed into time-frequency representations using continuous wavelet transforms (CWTs). Radiomic features were extracted to develop a deep learning classifier. The model achieved a macro-averaged AU-ROC of 0.88 and an overall accuracy of approximately 79%, demonstrating the feasibility of noninvasive, functional biomarkers for detecting early-stage dementia and differentiating it from healthy aging.
Taken together, these studies demonstrate new ways to extend the clinical utility of MEG across distinct neurological applications. This work expands MEG access to patient populations previously considered unsuitable, such as those with implanted RNS devices, by introducing an automated artifact rejection framework. It also introduces a fully automated, image-based spike detection pipeline to reduce the time burden on clinicians and improve consistency in MEG interpretation. Finally, it presents a novel application of deep learning and time-frequency analysis for non-invasive detection of cognitive decline, highlighting MEG's potential as a functional biomarker in dementia. Future work will focus on validating these methods in larger, multicenter cohorts, exploring longitudinal trajectories of neurodegenerative and epileptogenic processes, and integrating additional neuroimaging and molecular biomarkers. These advances offer promising avenues to improve clinical decision-making and optimize patient outcomes in epilepsy and dementia care.
Keywords
Epilepsy, Magnetoencephalography, Dementia, Signal processing, Deep learning, Mild cognitive impairment
Disciplines
Biological Engineering
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Recommended Citation
Askari, Pegah, "STRATEGIES FOR ENHANCED MEG DATA ANALYSIS IN CLINICAL PRACTICE AND EMERGING FRONTIERS" (2025). Bioengineering Dissertations. 201.
https://mavmatrix.uta.edu/bioengineering_dissertations/201