Virginia Tech - Wake Forest University School of Biomedical Engineering & Sciences
VTech Works (Virginia Tech)Not a member yet
122049 research outputs found
Sort by
LLM-Based Multi-Agent System and Simplicial Self-Supervised Learning Model for Regional Cancer Prevalence Estimation Using Satellite Imagery
No full textTraditional cancer rate estimations are often limited in spatial resolutions and lack considerations of environmental factors. Satellite imagery has become a vital data source for monitoring diverse urban environments, supporting applications across environmental, socio-demographic, and public health domains. However, while deep learning (DL) tools, particularly convolutional neural networks, have demonstrated strong performance in extracting features from high-resolution imagery, their reliance on local spatial cues often limits their ability to capture complex, non-local, and higher-order structural information. To overcome this limitation, we propose a novel LLM-based multi-agent coordination system for satellite image analysis, which integrates visual and contextual reasoning through a simplicial contrastive learning framework (Agent- SNN). Our Agent-SNN contains two augmented superpixel-based graphs and maximizes mutual information between their latent simplicial complex representations, thereby enabling the system to learn both local and global topological features. The LLM-based agents generate structured prompts that guide the alignment of these representations across modalities. Experiments with satellite imagery of Los Angeles and San Diego demonstrate that Agent-SNN achieves significant improvements over state-of-the-art baselines in regional cancer prevalence estimation tasks.Published versio
Collaborative Multimodal XR-based Training Environments for Collocated Medical Teams
No full textCollaborative Extended Reality (XR) systems hold growing promise as training platforms in domains that demand high levels of coordination, communication, and shared situational awareness. However, the current landscape of XR-based training tools remains predominantly focused on individual skill development, with limited support for realistic, synchronous collaboration among physically collocated teams. Furthermore, existing evaluation methods often rely on individual performance metrics and fail to capture the nuanced dynamics of teamwork. This dissertation addresses these critical gaps by proposing both design and evaluation frameworks tailored for collaborative XR training environments.
The research is structured around two core research questions. First, it investigates how XR training systems can be designed to support realistic, high-fidelity collaboration among collocated professional teams. Through naturalistic observations, stakeholder interviews, and iterative prototyping, the study identifies key design factors and formulates a set of principles that inform the development of a task-driven XR training simulator. Second, it introduces a theoretically grounded, multimodal, user-centered evaluation evaluation based on the Distributed Cognition Theory. This framework integrates behavioral, communicative, and perceptual data to assess team-level performance in XR, extending beyond traditional task metrics to include communication flow, role coordination, and temporal organization.
Together, the design and evaluation components contribute to a robust methodological pipeline for advancing Collaborative XR systems. The work not only bridges theoretical and practical gaps in XR training but also lays the groundwork for scalable, evidence-based tools that better reflect the realities of team-based performance in complex environments. Through these contributions, the dissertation advances the state of the art in collaborative immersive training and supports the development of next-generation XR platforms for real-world readiness.Doctor of PhilosophyExtended Reality (XR) technologies are becoming powerful tools for professional training in fields like medicine, emergency response, and technical operations. These immersive systems allow people to practice complex tasks in realistic, risk-free environments. However, most XR training tools today are built to train individuals, not teams. In the real world, professionals often work in groups where success depends not just on technical ability, but on how effectively team members communicate, collaborate, and make decisions together in real time.
This research aims to make XR training more collaborative and more realistic for teams who work physically together—what we call collocated teams. The first goal is to design XR environments that support natural teamwork by enabling team members to interact, coordinate, and share information as they would in real-life scenarios. This includes understanding what makes teamwork effective and turning those insights into design guidelines for building better XR simulators.
The second goal is to develop smarter ways to evaluate the collaborative XR settings through user data. Most current evaluation tools only measure individual performance, which misses key teamwork behaviors like coordination, leadership, or shared awareness. This research proposes a new framework for evaluating collaborative XR systems and the teams training in it by combining different kinds of data like communication, movement, and user performance to get a better picture of how well does the developed system support the team training.
The outcomes of this work will help create XR training systems that are not only more effective but also more aligned with the realities of how professionals operate in high-pressure environments.
Overall this research can lead to better team preparedness, improved outcomes, and enhanced safety across critical industries that rely on expert collaboration
Sparsity-aware Kernel Selection for Edge-Connected Jaccard Similarity in Graph Datasets
No full textPerformance of graph algorithms often depends both on the underlying hardware architecture and on structural properties of the input graph. Optimizations that deliver high performance on one class of graphs, such as hypersparse graphs with low average degree, can degrade performance on other classes, for example denser graphs with high average degree. In this work, we investigate sparsityaware GPU kernel selection for computing the Jaccard similarity index, a measure of neighborhood overlap in graph datasets. In our kernel selection approach, we use the vertex-centric Jaccard similarity implementation from the cuGraph library as the baseline and include both vertex- and edge-centric variants of this kernel, with set-intersection algorithms varying between two-pointer linear search, binary search, and adaptive dynamic search.
We use 80 real-world graphs in our evaluation with variation in average degree, maximum degree, Gini index, and average intersection cost. A random forest classifier, trained on a subset of these graphs on an NVIDIA A100 GPU, achieves 88.8% inference accuracy in predicting the fastest kernel. Kernels selected by the classifier achieve a 4.37× mean speedup over the vertex-centric cuGraph baseline from NVIDIA.Published versio
Healthcare in Wartime Ukraine
No full textThis internet publication shares reflections from my recent trip to Ukraine (2025), where I conducted interviews and gathered data in Ivano-Frankivsk, Lviv, and several surrounding communities. I spoke with health care providers and patients to understand their perspectives on the state of Ukraine’s health care system, the impact of ongoing reforms, and the challenges of accessing essential resources during wartime. This research was approved by the Virginia Tech Institutional Review Board (IRB #25-028) and the Ukrainian Institute of Public Health Policy (IRB #2025-2). While peer-reviewed publications are forthcoming, what follows offers an early glimpse into these findings.Published versio
Numerical Simulations of Supercavitating Propellers and Hydrofoils
No full textA systematic investigation is conducted to compare and evaluate the hydrodynamic performance and cavitation pattern prediction capabilities of URANS and BEM solvers for supercavitating propellers (SCPs) and hydrofoils (SCHs) with non-conventional sectional profiles. The previously developed BEM at the VT Innovative Ship Design Lab (VT-iShip) is extended upon to allow for supercavitating profiles with truncated trailing edges (TE). Both the BEM and the URANS solutions are validated against experimental data for a range of operating conditions and their discrepancies from experimental trends are quantified. The predicted solutions from both methods closely overlap with experimental data and contain an experimental error in the order of 10% for a large range of operating conditions. Some novel contributions in this study include various theoretical developments that allow the BEM to now consider a large number of supercavitating conditions for supercavitating profiles with truncated TEs. The theoretical modifications to the BEM algorithms are developed into BEM executable frameworks for 2D and 3D SCHs and SCPs. Moreover, the novel contributions also include the identification of select URANS turbulence and cavitation models to most accurately predict supercavitating propeller and hydrofoil performance parameters and cavitation patterns. These solutions are compared with BEM solutions. A design space evaluation of each method is also conducted along with various experimental validation studies to evaluate the robustness of the BEM algorithm. A comparison of their required computational and time-based resources reveals that the methods can be used in a complementary manner for a large range of operating conditions when evaluating designs in the supercavitating regime.Doctor of PhilosophyThere is an ever-growing demand to increase the maximum attainable speed for high speed marine crafts coupled with the use of high-powered outboard engines. This is through the development and innovation of marine propulsors that are capable of efficiently operating in the supercavitating flow regime. The design optimization of these innovative and unconventional propellers solely through experimentation or high-fidelity numerical solvers, such as the Reynolds Averaged Navier Stokes Equations (RANSE), can be extremely costly and time consuming. Moreover, the capabilities of the modern day RANSE solvers in resolving complex, multiphase, and separated flow are still being explored. An alternative method for cavitating flow prediction includes the low order, Boundary Element Method (BEM) that needs to be experimentally validated for supercavitating conditions. This study systematically explores the feasibility of utilizing the BEM and Unsteady RANSE solvers in a complementary manner to predict the performance of non-conventionally shaped supercavitating propellers (SCP) and hydrofoils (SCH) in 2D and 3D. The study explores the capabilities and limitations of the two methods in a wide range of operating conditions. It also documents the theoretical developments made to the 2D and 3D BEM algorithm to improve its compatibility with profiles featuring truncated trailing edges (TE). These algorithms are then developed into executable frameworks. The novel contributions in this study include the theoretical developments made to the BEM for SCHs and SCPs, the identification of select physics models to most accurately predict the SCH and SCP hydrodynamic performance and cavitation patterns, their comparison with the BEM solutions, and an evaluation of the design space to which the BEM algorithm is applicable and reliable
Lubricants
No full textThe paper proposes a pavement safety index, the estimated available friction at the expected travel speed, FRS(v), to model the composed effect of low-slip speed friction and macrotexture on roadway crashes. This index seems to capture the relative contributions of microtexture and macrotexture across different operating speeds. Speed-dependent available friction at 40, 55, and 70 mph was estimated using the speed-correction procedure in ASTM E1960-07 and integrated into Safety Performance Function (SPF) development. Comparison of the resulting SPF models suggests that FRS values corresponding to typical operating speeds can capture the combined influence of SFN (40) and macrotexture on expected crashes for freeways and rural two-lane, two-way highways. For freeways, the estimated available friction at 70 mph (FRS113) produced the most appropriate SPF, evidenced by the lowest AIC. For rural two-lane, two-way highways, the estimated available friction at 40 mph (FRS65) resulted in the lowest AIC value, consistent with the typical operating speeds on these facilities. In contrast, none of the speed-specific friction estimates produced satisfactory model performance for urban and suburban arterials, likely due to the wide variation in traveling speeds and geometric characteristics on these facilities. The applicability of the proposed metric was demonstrated through the development of illustrative investigatory friction levels based on observed crash data, and the identification of candidate roadway segments for friction improvement interventions, and the estimation of the corresponding return on investment for these interventions.Published versio
Stories of the Wind
No full textStories of the Wind is an audiovisual performance exploring various media to tell a story, integrating media at the intersection of visual arts and music, leveraged by technology. Different materials and technologies coexist as pieces of an audiovisual performance, with images, sound objects and interactive works. The production of this work was informed by artistic-scholarship, which involved the combination of aesthetic education and aesthetic experience with research and analysis in the process of artistic and academic creation.
This project was meant to be exhibited in the Cube, at the Moss Arts Center, at Virginia Tech. Because of the Covid-19 pandemic, it was not possible to present the project in the space that it was created for, so a video adaptation was made to be submitted for the thesis defense. The video submitted as the thesis project pandemic adaptation can be seen through the following link: ​https://www.youtube.com/watch?v=dH8ce9KO41wandt=50sDoctor of PhilosophyStories of the Wind is an audiovisual performance telling a story using visual arts and music. It was created to be performed in the Cube, at the Moss Arts Center, at Virginia Tech.
Because of the Covid-19 pandemic, it was not possible to perform the project. A video adaptation was made instead, which can be seen through the following link: https://www.youtube.com/watch?v=dH8ce9KO41wandt=50
Addiction
No full textAims: We examined the feasibility and acceptability of pairing portable breathalyzers to assess field alcohol use with mobile ecological momentary assessment (EMA) to assess intimate partner violence (IPV; psychological, cyber, physical, and sexual aggression) perpetration and victimization among undergraduates who drink heavily and were recently aggressive. Design, Setting, and Participants: We assessed EMA/breathalyzer completion rates, drinking captured via breathalyzer versus self-report, number of IPV events captured, procedural acceptability, and reactivity to assessment. Sex differences were examined. Undergraduates aged 18-25 (N = 103; M age = 21 years, SD = 2.0; 52% women; 80.6% heterosexual; 64.1% white; 93.2% non-Hispanic) recruited from a large Mid-Atlantic university in the United States completed a baseline survey then a 30-day EMA wherein they were prompted to complete one morning and three evening surveys (7PM, 9PM, 11PM) daily. After each evening survey, participants were prompted to submit a breath alcohol content (BrAC) sample to a breathalyzer linked to surveys. Participants could self-initiate surveys after drinking or IPV outside of assessment periods. Afterward, participants completed an exit survey. Measurements: Outcome variables were self-reported alcohol use and IPV assessed via EMA surveys, and BrAC assessed via breathalyzer. Self-reported procedural acceptability was assessed in the exit survey. Reactivity to assessment was assessed by analyzing daily trends in IPV and drinking by sex using generalized linear mixed effects models. Findings: Participants completed 80% of surveys and responded to 91% of breathalyzer prompts. BrAC was captured in 89.4% of self-reported drinking events, 91.4% of self-reported non-drinking events, and 95.8% of IPV events, with greater responsiveness to breathalyzer prompts as the evening progressed despite increasing intoxication. More IPV events were captured during evening and event triggered (358 combined total events) than morning surveys (245 events). Results were comparable across women and men. Each additional study day was associated with modest declines in odds of experiencing any IPV (odds ratio [OR] = 0.95, 95% confidence interval [CI]: 0.94-0.97, p < 0.001), IPV perpetration (OR = 0.94, 95% CI: 0.92-0.96, p < 0.001), IPV victimization (OR = 0.97, 95% CI: 0.96-0.99, p = 0.004), any drinking (OR = 0.99, 95% CI: 0.98-1.00, p = 0.01), and positive BrAC readings (OR = 0.99, 95% CI: 0.98-1.00, p = 0.052), suggesting minimal reactivity to assessment. Participants reported high overall satisfaction with study components. Conclusions: Pairing ecological momentary assessment with portable breathalyzers to capture data on drinking and intimate partner violence across 30 days among US undergraduates who were previously aggressive and who drink heavily appears to be both feasible and acceptable.This work was supported by grant R21AA030858 from the National Institute on Alcohol Abuse and Alcoholism (NIAAA) awarded to the first author. The content is solely the responsibility of the author and does not necessarily represent the official views of the NIAAA, or the National Institutes of Health.Accepted versio
Tourism Management
No full textThis study examines the effectiveness of framing strategies—diagnostic, prognostic, and motivational—in marketing regenerative tourism, drawing on the framing perspective within social movement theory and the concept of frame resonance. Using a sequential mixed-methods approach, Study 1 features 23 semi-structured in-depth interviews with Destination Management Organization (DMO) leaders and regenerative practitioners. Participants report that positive messaging combined with clear calls to action resonate strongly with tourists. Study 2 employs a between-subjects online experiment with respondents from the United States, the United Kingdom, and New Zealand to test these insights. Results reveal that framing strategies and frame resonance significantly influence tourists’ attitudes toward and intentions to participate in regenerative tourism. Theoretically, this research advances the application of framing strategies, frame resonance, and social movement theory in the context of regenerative tourism. Practically, it recommends that DMOs apply prognostic (solution-oriented) and motivational (call to action) framing that emphasizes experiential benefits and provides actionable steps to engage target audiences effectively.Accepted versio
Essays on the Economics of Food and Agricultural Technology
No full textTechnological adoption in food and agricultural systems has long been recognized as a major driver of productivity growth, structural transformation, and welfare improvement. This dissertation examines technology adoption and its welfare implications across nutrition and agricultural production.
The first chapter evaluates a fortified rice intervention implemented through a randomized school feeding program in Cambodia. Using panel data with pre- and post-intervention measurements, the analysis estimates the impacts of fortified rice on children's nutritional outcomes, micronutrient biomarkers, and cognitive performance. The results show significant reductions in zinc deficiency and improvements in selected biomarkers, with effects varying by rice formulation and micronutrient composition. Significant heterogeneity is observed by child characteristics, including sex, age, and baseline nutritional status. Beyond the probability of having deficiency, the intervention also reduces the zinc deficiency gap and severity.
The second chapter assesses the cost-effectiveness of three fortified rice formulations based on micronutrient composition, using Difference-in-Differences estimates from the first chapter as measures of effectiveness to construct Incremental Cost-Effectiveness Ratios. While the type of fortified rice with higher contents of serum zinc is the most effective formulation for zinc-related outcomes, it lies in the "uncertain" region of the cost-effectiveness plane due to its higher incremental cost, highlighting the importance of explicit Willingness-to-Pay thresholds for adoption decisions.
The final chapter analyzes the determinants and market impacts of controlled environment agriculture (CEA) in U.S. vegetable production. Using a random forest model, the chapter identifies key drivers of adoption, including temperature, electricity prices, and financial capacity, and predicts state-level adoption rates. These predicted adoption rates are then incorporated into an equilibrium displacement model to simulate CEA expansion and electricity price shocks. The results indicate that CEA expansion modestly lowers prices, increases quantities, raises consumer surplus, and generates heterogeneous welfare effects across production systems. In addition, CEA adopters operating energy-intensive systems are more sensitive to changes in electricity prices.Doctor of PhilosophyTechnology adoption in food and agricultural systems plays a major role in improving productivity, transforming food systems, and enhancing economic well-being. This dissertation examines how technology adoption in nutrition and agriculture affect nutritional outcomes and market outcomes.
The first chapter evaluates a school feeding program in Cambodia that provided fortified rice to children. The results show that fortified rice improved nutritional outcomes, particularly by reducing zinc deficiency, and that the impacts varied by child characteristics, including sex, age, and baseline nutritional status. The program also reduced the zinc deficiency gap and severity among children who remained deficient.
The second chapter examines whether these nutritional improvements were cost-effective. Among three fortified rice formulations, the option with higher zinc content delivered the strongest improvements but at a higher cost, making adoption decisions dependent on policymakers' willingness to pay for improved nutritional outcomes.
The final chapter focuses on U.S. vegetable production and the adoption of controlled environment agriculture (CEA), such as hydroponics. The analysis shows that climate conditions, energy prices, and financial capacity strongly influence adoption. Expanding these systems can lower vegetable prices and benefit consumers, but the effects differ across farmers depending on whether they adopt CEA. In addition, producers adopting CEA systems are more affected by changes in electricity prices