ORCID Identifier(s)

ORCID 0000-0002-1850-5299

Graduation Semester and Year

Fall 2025

Language

English

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Biomedical Engineering

Department

Bioengineering

First Advisor

Dr. Christos Papadelis

Second Advisor

Dr. George Alexandrakis

Third Advisor

Dr. Steven Stufflebeam

Fourth Advisor

Dr. Hanli Liu

Abstract

Drug-resistant epilepsy (DRE), which affects approximately 30% of individuals with epilepsy, presents a significant clinical challenge, as seizure control cannot be achieved with anti-seizure medications. The most effective treatment, epilepsy surgery, relies on the accurate delineation of the epileptogenic zone (EZ), the area of the brain capable of generating seizures. However, surgical success is often constrained by current methods for EZ localization, which typically depend on capturing spontaneous seizures to define the seizure onset zone (SOZ) using intracranial electroencephalography (iEEG). This approach requires prolonged invasive monitoring, is costly, and carries potential risks and discomfort for patients, highlighting the need for reliable interictal biomarkers that can identify the EZ without waiting for seizures to occur.

To overcome the limitations of seizure-based localization, research has increasingly focused on interictal biomarkers, abnormal electrophysiological events that occur between seizures and provide information about epileptogenic brain activity. Among these biomarkers, interictal spikes (1–70 Hz) and high-frequency oscillations (HFOs) have emerged as the most promising indicators of the EZ. HFOs are further categorized into ripples (80–250 Hz) and fast ripples (250–500 Hz), which are believed to reflect hypersynchronous neuronal firing within epileptic networks. Spikes and HFOs are frequently observed in regions associated with seizure generation, and higher rates and powers of these events have been correlated with epileptogenic tissue and favorable surgical outcomes when resected. However, these interictal biomarkers do not occur in isolation: both spikes and HFOs often propagate across spatially distributed brain regions in a temporally ordered manner, resembling the spread of seizures. Mapping these spatiotemporal propagations provides valuable insight into the structure of epileptic networks, where propagation onset regions are more likely to represent primary generators, while propagation spread regions reflect secondary involvement. Characterizing these propagation dynamics may therefore improve the precision of EZ delineation beyond conventional rate- or power-based measures.

While invasive iEEG recordings provide direct access to epileptogenic brain regions, there remains a strong clinical need for non-invasive interictal biomarkers that can guide presurgical planning in a broader range of patients. In clinical practice, interictal spikes detected with magnetoencephalography (MEG) and electroencephalography (EEG) are the most widely used non-invasive biomarkers for localizing epileptogenic activity, particularly during the first phase of presurgical evaluation in pediatric patients with DRE to guide iEEG electrode placement. Source localization of these spikes through magnetic and electrical source imaging (MSI and ESI) enables clinicians to estimate the cortical generators of epileptiform discharges, infer the location of the EZ, and support critical surgical decision-making. However, the accuracy of MEG and EEG source localization depends heavily on factors such as source depth, orientation, tissue conductivity, and background noise. To ensure their reliability, realistic physical head phantoms have become indispensable tools for validating source localization algorithms. Such phantoms provide controlled, repeatable environments with known ground-truth source locations, allowing systematic evaluation of how well MSI and ESI can localize deep and superficial epileptic sources under different noise conditions.

In this thesis, we systematically mapped the spatiotemporal propagation of interictal spikes, ripples, and fast ripples in patients with DRE, quantifying their spatial and temporal features. Through this analysis, we identified a novel biomarker, the spike–ripple onset overlap (SRO) zone, representing the region where spike and ripple propagations share a common onset generator. Our findings demonstrate that the SRO zone corresponds closely to the EZ, showing higher specificity than conventional biomarkers, and that its resection is associated with favorable surgical outcomes. To further enhance objectivity and reproducibility, we developed a machine learning framework that integrates propagation properties with additional spatial and temporal features. This data-driven approach aims to automate the delineation of the EZ, reducing reliance on subjective visual interpretation and minimizing human error.

Building on the need for validation of non-invasive source localization methods, we designed and fabricated a realistic three-layer pediatric head phantom, comprising brain, skull, and scalp compartments with physiologically appropriate conductivities. The phantom contained multiple implanted deep dipolar sources in regions commonly involved in different forms of epilepsy and was driven with real interictal spike waveforms derived from iEEG recordings of a patient with DRE. MEG and EEG data were then acquired under controlled noise conditions, and source localization was performed using two clinically established approaches: the equivalent current dipole (ECD) and dynamic statistical parametric mapping (dSPM). By comparing estimated versus true source locations across varying signal-to-noise ratios, we quantitatively assessed the reliability of MEG and EEG source localization for deep epileptogenic sources.

Together, these invasive and non-invasive investigations converge toward a unified goal: developing robust, quantitative, and minimally invasive methods for accurate delineation of the EZ. By combining advanced propagation modeling, machine learning integration, and empirical validation through realistic phantom experiments, this work contributes to the establishment of reproducible, clinically applicable biomarkers that enhance presurgical planning and ultimately improve surgical outcomes in patients with DRE.

Keywords

EEG, MEG, electric source imaging, magnetic source imaging, ESI, MSI, Phantom, DRE, Interictal Biomarkers, Machine Learning

Disciplines

Bioelectrical and Neuroengineering | Biomedical Devices and Instrumentation | Other Biomedical Engineering and Bioengineering

Download

Available for download on Friday, December 31, 2027

Share

COinS