Award Date

5-1-2025

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Interdisciplinary Programs

First Committee Member

Brach Poston

Second Committee Member

Graham McGinnis

Third Committee Member

Mark Guadagnoli

Fourth Committee Member

Michael Lee

Number of Pages

185

Abstract

Parkinson’s disease (PD) affects over 10 million people and its prevalence is projected to double by 2040 worldwide. Current pharmacological or surgical options provide only symptomatic relief. Anodal transcranial direct‑current stimulation (tDCS) of the primary motor cortex (M1) is a promising, inexpensive adjunct treatment modality that could be efficacious in PD. However, its optimal stimulation parameters remain unknown, and the rapidly expanding trial literature makes it challenging to interrogate manually. To address these issues, we built a large‑language‑model pipeline that automatically extracted, cleaned and tagged every PD and tDCS record on ClinicalTrials.gov. This resulted in 64 trials, double the count of a standard keyword search, with perfect inter‑model agreement on core stimulation parameters. In addition, two experimental studies were completed to compare different tDCS parameters of stimulation on motor skill, transfer of motor skill, and cortical excitability in PD.Chapter 1 provides a comprehensive overview of the history, evolution, mechanisms, theories, and specific applications of non-invasive brain stimulation, and introduces relevant motor learning principles to establish a foundation of essential knowledge. The chapter is intentionally written in a clear, non-technical style to ensure that readers can grasp the technical content presented in chapters two through four, regardless of their initial understanding of brain stimulation and motor learning. Chapter 2 aimed to determine whether a parallel‑LLM extraction pipeline (GPT‑4, Claude 3.5 Sonnet, Llama 3.1 70B) can retrieve and structure ClinicalTrials.gov data on tDCS in PD more completely and efficiently than conventional keyword‑based searches. The secondary purpose was to compare the three LLMs’ relative strengths and limitations in parsing clinical‑trial protocols, quantify their agreement on key elements (e.g., stimulation parameters, anatomical targets, treatment duration), and identify where human oversight is still required. The study method was an automated, API‑driven, single‑dataset, parallel‑processing implementation in which sixty‑four PD‑tDCS trials (identified with API queries plus regex filtering) were fed in parallel to the three LLMs under an identical, JSON‑constrained prompt. Each model’s output was schema‑validated and logged; discrepancies were analyzed with percent agreement, Cohen’s κ, ICC, and error metrics. The final outcomes included total trials captured versus baseline search (64 vs 28), inter‑model reliability for simple (κ > 0.90) and complex (κ ≈ 0.60) attributes, and numeric‑parameter concordance (intensity ICC = 1.0; duration ICC = 0.35). Pairwise comparisons showed substantial agreement among language models, with the highest concordance seen for identifying brain stimulation presence (97.9%) and non-invasive classifications (96.8%); reliability varied across models and extracted parameters. While large language models exhibit strong potential for automating trial protocol analysis, their varied performance on complex extraction tasks highlights the continued need for human oversight and multi-model ensemble approaches to ensure accuracy and reliability. Chapter 3 investigated the impact of tDCS applied before versus during motor practice on motor skill acquisition in PD. The secondary purpose was to determine the influence of tDCS on the transfer of motor skill in PD. The study employed a single-blind, SHAM-controlled, within-subjects design, and 12 individuals with PD completed a SHAM condition, a tDCS before motor practice condition (BEFORE), and a tDCS during motor practice condition (DURING) in three separate experimental sessions. The practice task was a complex visuomotor isometric precision grip task (PGT). In addition, motor skill transfer tasks and transcranial magnetic stimulation (TMS) measures of cortical excitability were performed pre- and post- each practice and stimulation period in the three conditions. The force error in the PGT was significantly lower in the BEFORE condition compared to the SHAM condition (p = 0.031). Force error was also lower in the DURING condition compared to the SHAM condition, but this difference was not statistically significant (p = 0.15). Furthermore, neither tDCS condition significantly enhanced performance in any of the motor skill transfer tasks. Similarly, there was no significant effect of tDCS on measures of cortical excitability. Collectively, these results indicate that tDCS of M1 applied before practice enhances motor skill acquisition in PD, but tDCS does not lead to significant increases in the transfer of motor skill or cortical excitability. Finally, in Chapter 4, the primary purpose of this study was to determine the influence of tDCS intensity on motor skill acquisition in individuals with PD. The secondary purpose was to examine the effects of tDCS intensity on motor skill transfer in PD. The study utilized a SHAM-controlled, single-blind, within-subjects crossover experimental design. A total of 15 individuals with PD performed three experimental sessions (SHAM, 1 mA tDCS, and 2 mA tDCS conditions) that involved practice of a complex visuomotor isometric precision grip task (PGT). Furthermore, assessments of motor skill transfer and transcranial magnetic stimulation (TMS) metrics of cortical excitability were administered prior to and following each practice and stimulation session in all three experimental conditions. Although, the force error in the PGT was lower in both the 1 mA and 2 mA tDCS conditions compared with the SHAM condition, a one-way ANOVA revealed that these differences failed statistical significance (p = 0.075; ηp2 = 0.169). Similarly, transfer task motor performance was not significantly improved and cortical excitability was not increased following either of the two tDCS conditions compared to the SHAM condition. The findings of this study collectively suggest that acute single session applications of tDCS at two commonly used stimulation intensities do not produce considerable enhancements in motor skill acquisition, motor skill transfer, or cortical excitability in PD.

Keywords

Clinical trial data extraction; Computational neuroscience; Large language models; Motor skill acquisition; Parkinson's disease; Transcranial direct current stimulation (tDCS)

Disciplines

Artificial Intelligence and Robotics | Computer Engineering | Medical Neurobiology | Neuroscience and Neurobiology | Neurosciences | Social and Behavioral Sciences

File Format

pdf

File Size

2300 KB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Rights

IN COPYRIGHT. For more information about this rights statement, please visit http://rightsstatements.org/vocab/InC/1.0/

Download

Available for download on Monday, May 15, 2028


Share

COinS