Author ORCID Identifier

https://orcid.org/0000-0001-8219-5101

Defense Date

2024

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Mechanical and Nuclear Engineering

First Advisor

Ravi L. Hadimani

Abstract

Transcranial Magnetic Stimulation (TMS) is a safe, effective, and non-invasive therapy for treating several psychiatric and neurological disorders. TMS is Food and Drug Administration (FDA) approved treatment and is commonly applied to patients who do not respond to medications for the treatment of clinical depression, smoking cessation, obsessive-compulsive disorder and migraine. Recently, there has been an increase in the development of electromagnetic neuromodulation techniques targeted at enhancing the effectiveness of TMS devices for the treatment of mental diseases. In TMS stimulation, focality is an important factor which determines the specificity of the pulses induced in different brain tissues. The electromagnetic pulses must be confined to specific regions of the brain and avoid stimulation of the surrounding areas. This thesis focuses on two complementary research avenues aimed at enhancing TMS therapy: the development of advanced TMS coils for small animal models and the application of machine learning to predict TMS-induced electric field (E-field) distributions efficiently and accurately.

The first phase of this research involved the design and development of a novel focal TMS coil tailored for small animal models. Small animal studies are critical for understanding cortical connectivity and exploring neuromodulation techniques before scaling them to human applications. However, the use of commercially available coils designed for human brains poses significant limitations when applied to smaller animals due to their broad stimulation fields. To address these issues, this study systematically investigated innovative coil configurations and advanced soft magnetic materials. A novel Parabolic Ferromagnetic Core (PFC) coil was developed, featuring a ferromagnetic iron-cobalt-vanadium alloy core paired with a pyrolytic graphite diamagnetic plate. The PFC coil demonstrated significant improvements in focality, producing highly localized stimulation areas as small as 1 mm², as confirmed through extensive finite element analysis (FEA) simulations using Sim4Life and ANSYS Maxwell. The inclusion of the diamagnetic plate further enhanced focality by redirecting magnetic flux to the target region, minimizing unintended activation of adjacent tissues.

Experimental validation of the PFC coil was conducted at the Richmond Veterans Affairs Medical Center in collaboration with Dr. Mark Baron’s laboratory. The coil was tested on rats, including recording motor-evoked potential (MEP) responses to evaluate its ability to stimulate motor cortex regions (M1 and M2) precisely. The results revealed superior E-field distribution and focality compared to other coils, highlighting its potential for future scaling to human applications. By addressing the limitations of current coils, the PFC coil represents a significant advancement in achieving precise neuromodulation, paving the way for its use in treating focal neurological disorders such as Parkinson’s disease and dystonia.

The second phase of this research explored the application of deep convolutional neural networks (DCNNs) to predict TMS-induced (E-fields) in real time. The model leveraged anatomical variations derived from T1- and T2-weighted magnetic resonance imaging (MRI) scans to accurately estimate the maximum induced E-field on targeted regions of the brain. Trained using 231 head models, the DCNN achieved peak training and validation Peak signal to noise ratios (PSNRs) of 34.77 dB and 29.08 dB, respectively, and corresponding MSEs of 3.335 × 10⁻⁴ and 1.237 × 10⁻³. This approach positions the DCNN as a computationally efficient alternative to time-intensive FEA methods, significantly enhancing the feasibility of real-time E-field predictions for neuromodulation research.

Then, we improved this model by developing a DCNN capable of real-time 3D E-field predictions using anatomical and E-field 3D image stacks. The model, trained on nonimpaired head models, demonstrated promising results when applied to structurally abnormal cohorts, such as mild traumatic brain injury (mTBI) patients. The U-net-based architecture effectively captured intricate E-field dynamics across diverse brain structures, highlighting its adaptability for optimizing TMS protocols. These findings underscore the model’s utility as a precise tool for both clinical applications and neuromodulation research.

Finally, the integration of MRI-derived anatomical features,electroencephalogram (EEG) data, and E-field simulations in the last phase of this study introduced a novel approach to predicting resting motor threshold (RMT). By incorporating predictors such as scalp-to-cortex distance, gray matter volume, and novel features like skull and skin volume percentages, thesupport vector machine (SVM) model achieved robust predictive performance. This phase underscores the importance of accounting for non-linear feature relationships in personalized TMS protocol development. Together, these efforts advance TMS as a precision therapy, providing a pathway to improved neuropsychiatric care through tailored interventions.

This dissertation presents significant advancements in TMS technology and methodology, spanning coil design, machine learning-based E-field prediction, and personalized therapy development. By addressing both the technical and clinical challenges of TMS, these studies contribute to the broader goal of establishing TMS as a precision neuromodulation tool with applications in both research and clinical settings.

Rights

© Mohannad Tashli

Is Part Of

VCU University Archives

Is Part Of

VCU Theses and Dissertations

Date of Submission

12-10-2024

Download

Share

COinS