Dissertation defense schedule

EchoBrain: AI-Enhanced EEG and Infrasound Analysis for Comprehensive Brain Function Assessment through Cognitive Signals

Neurological disorders are among the leading causes of death and disability worldwide, yet the tools to observe brain function remain costly, cumbersome, and clinic-bound: a single electroencephalogram (EEG) in the United States averages roughly US$972 and requires trained staff and gel electrodes. This dissertation develops EchoBrain, an artificial-intelligence-enhanced framework that assesses brain function from cognitive signals, namely conventional scalp EEG together with a novel acoustic view of cerebral activity (the infrasound or resonance frequency produced by cerebral blood flow), and moves toward an accessible, low-cost, non-invasive alternative to traditional EEG.

The work spans four aims. Aim 1 introduces an interpretable, chaos-theory-based method for detecting attention from EEG (time delay, embedding dimension, and correlation dimension), framed by a survey of nonlinear EEG techniques. Aim 2 delivers a real-time brain-computer interface (BCI) that classifies no-, single-, and consecutive two-blink events, reaching 89% accuracy with classical models and 98.67% recall with a YOLOv8 detector. Aim 3 advances the central hypothesis that blood-flow infrasound carries brainstate information, implementing the NeuroEcho pipeline that maps acoustic windows to EEG band-states across ten participants and fifty sessions. Aim 4 frames the system within agentic AI for BCIs and validates an EchoBrain prototype with ten participants performing blink-based yes/no communication, attaining 97% mean accuracy and a System Usability Scale score of 94.25. Together the aims demonstrate an end-to-end path from interpretable EEG analysis to a validated multimodal brain-assessment system; the dissertation also documents current limitations and outlines future work, including generative waveform reconstruction, an earplug infrasound device, and larger clinical cohorts.

ENGINEERING THE MICROENVIRONMENT OF SINGLE-ATOM CATALYSTS THROUGH SURFACE MODIFICATION FOR CROSS-COUPLING REACTIONS

Single-atom catalysts (SACs) have emerged as a promising class of heterogeneous catalysts capable of bridging the gap between homogeneous and heterogeneous catalysis. By distributing isolated metal atoms across a solid support, SACs offer uniform active sites, near-total metal utilization, straightforward separation and recyclability. These features make SACs compelling candidates for reactions traditionally catalyzed by homogeneous transition-metal complexes, such as carbon–
carbon cross-coupling reactions. However, unlike homogeneous catalysts, SACs lack molecular ligands that can be rationally tuned to control the environment around the active metal center. This dissertation demonstrates that organic ligands bound to the support surface can be used to modify the microenvironment around the active sites and enhance SAC performance in cross-coupling catalysis.

Chapter 1 provides a brief review of the SAC field, covering its historical
development, synthesis and characterization techniques, and catalytic applications. Chapter 2 introduces the surface-modification strategy by coating Pd/CeO₂ SAC with benzoic acid-based ligands and systematically investigating their electronic properties. The study then evaluates the substituent effects on the aryl bromide and organoboron coupling partners for both uncoated and coated Pd/CeO₂. We find that coating the support with benzoic acid-based molecules increases the yield of Suzuki reaction several-fold, an effect that holds across a broad range of reactants and substituted benzoic acids.

Chapter 3 extends this surface-modification strategy to Sonogashira coupling, demonstrating that carboxylate-based ligands similarly enhance the activity of Pd/CeO₂ for this distinct reaction class. The chapter first establishes how the effect of the organic layer varies with solvent environment, then examines several possible origins of the activity enhancement by comparing aromatic carboxylate ligands with varying tether lengths and nonaromatic carboxylate ligands. Notably, both aromatic and nonaromatic ligands improve Sonogashira coupling activity regardless of tail group identity, suggesting that the carboxylate anchoring group contributes to the observed enhancement.

This is the first demonstration in the literature that support-bound carboxylate ligands can enhance the activity of SACs in carbon–carbon cross-coupling reactions. This work opens a path toward rationally designed SACs in which support–ligand interactions are deliberately engineered to create tunable microenvironments around isolated metal
active sites.

Life-Cycle Management of Deteriorating Pipeline Networks

The secure transportation of energy commodities via pipelines is essential to sustaining modern life. In the United States alone, an extensive network of approximately three million miles of pipelines spans the nation’s geography, delivering fuels and feedstocks that underpin economic activity and public welfare. Failures in any part of the pipeline network can trigger devastating incidents, endangering human life, harming the environment, and imposing substantial economic losses. Hence, informed-management decisions are required to protect pipeline networks against these failures.

To make informed integrity-management decisions (e.g., inspection, repair, and replacement), operators increasingly rely on quantified failure consequences. Although many studies have used failurecost
consequences to determine optimal inspection intervals in corrosion-affected pipelines, the field still lacks a widely accepted, predictive model for the financial consequences of failure.

Cost-effective integrity management is fundamentally a life-cycle optimization problem that seeks to minimize expected total costs while meeting reliability targets. Achieving this goal requires a reliable prediction of degradation progression under uncertainty. Robust prognostics should explicitly model parameter uncertainty, measurement error, and model error, and then assimilate new inspection data to refine future predictions. Furthermore, since risk depends jointly on demand and capacity, decision-quality assessments must also incorporate accurate capacity models for recently deployed pipeline materials, such as polyethylene with detected defects under operating pressure. When reliability and risk are evaluated with calibrated models, infrastructure managers can deploy maintenance resources more effectively and align interventions with system-level performance and safety goals.

Across materials and threat mechanisms, pipelines operating in corrosive or otherwise degrading environments experience evolving structural states that can culminate in failure. Accurate prediction of failure time over the service life, especially under repeated inspections and potential repairs, is therefore central to planning inspection intervals, setting repair thresholds, and optimizing life-cycle strategies. To support these needs, a probabilistic analytical perspective is required to model how failure probability evolves for pipelines with growing defects, while accounting for inspection-triggered repair actions. Such an analytical approach enables probability-conserving propagation of risk through time and provides a computationally efficient alternative or complement to simulation-based strategy evaluation.

Despite progress, important gaps remain. First, widely used failure-cost assumptions often rely on coarse categorization, which can bias life-cycle optimization; Hence, a predictive, attribute-based consequence model is needed. In addition, defect growth and service-life prediction have not fully integrated inspection data and model error, limiting the operational value of monitoring data. Furthermore, for deteriorating pipelines, an analytical (non-simulation) framework that conserves probability across
inspection/repair nodes and constructs branch-wise failure probability is needed to support life-cycle management. Finally, while many frameworks have been proposed to optimize life-cycle management strategies, past approaches have relied on simplifying assumptions or on simulations that do not explicitly track the dependence between failure events and maintenance actions through an analytical approach in multi-segment pipelines.

In brief, the goal of this research is to (i) advance predictive, attribute-based models of failure cost consequence for use in life-cycle optimization; (ii) develop accurate time‑evolving deterioration models
and failure time prediction for pipelines by integrating inspection data; (iii) establish an analytical framework considering the repair impact on failure probability of defected pipelines, and (iv) propose an objective-based life-cycle management tool that optimizes inspection and repair policies for pipelines with defects, balancing reliability targets and total expected cost. Addressing these needs will enable more defensible, cost-effective integrity-management strategies across pipeline systems.

Dedicated breast X-ray computed tomography (breast CT) is a developing imaging modality that has potential as an alternative to mammography. Unlike traditional mammography, breast CT offers high-resolution, fully 3D anatomical detail, which aids in the detection of microcalfications and subtle lesions that are potential indicators of cancer. To be effective for screening, breast CT must limit radiation
exposure by reducing CT view angles. However, traditional image reconstruction methods yield noisy, artifacted images in this sparse-view CT setting.

This dissertation develops image restoration and reconstruction methods that address these limitations by integrating physics-informed iterative methods with machine learning tools, and a novel neural network training approach that aligns training objectives with clinically relevant signal detection tasks arising in breast CT.

To recover material-specific tissue maps from nonlinear dual-energy breast CT data, an unrolled estimator that embeds a learned network as both initialization and regularizer within a fixed number of iterations of a model-based reconstruction algorithm is proposed. On simulated dual-energy breast CT data, the unrolled approach achieves superior pixel-wise image quality at lower computational cost than full iterative reconstruction.

To address the misalignment between pixel-wise training losses and clinical signal detection tasks in breast CT, a novel training objective function, the signal promoter (SigPro) loss, is introduced. The SigPro loss embeds a synthetic binary signal detection task into the neural network training. Evaluated on a textured digital breast CT phantom, denoising networks trained with SigPro loss achieve higher signal detection performance than those trained with a more traditional pixel-wise loss.

To enforce data consistency while preserving task-informed image priors, a pre-trained SigPro denoising network is embedded into an iterative reconstruction algorithm. The resulting estimator achieves signal detectability performance approaching the theoretical limit, substantially outperforming stand alone denoising networks and a related estimator based on a pixel-wise loss trained denoiser.

To adapt the proposed image restoration methods to clinically realistic settings, a 3D fusion network is proposed and evaluated on high resolution anatomically realistic breast phantoms, providing proof-of-concept for extending task-informed deep learning based CT image restoration from 2D slices to 3D volumes.