Master's Theses

Permanent URI for this collectionhttps://hdl.handle.net/10161/2493

For access to theses created prior to 2006 and those not submitted electronically, please see: https://library.duke.edu/find/theses-dissertations.

Policies & procedures governing electronic theses & dissertations can be found on the graduate school website.

Browse

Now showing 1 - 20 of 1787

Access status: Open Access ,
Development and Validation of Tools for Virtual Imaging Trials of Ultrafast Breast MRI
(2026) Carrascosa, Roberto A
Ultrafast MRI is an advanced technique used for identification of breast lesions, consistent with cancer. It is a contrast-enhanced scan which utilizes several undersampling techniques, all of which can be applied to varying degrees. This level of complexity, coupled with the inability to repeatedly inject contrast into a patient. makes the scan difficult to optimize in the clinic. To address this challenge, we designed an MRI simulator and corresponding phantom which operates on fundamental MRI principles to evaluate temporal performance of various TWIST settings and provide clinical recommendations. Our simulations revealed complex interactions between GRAPPA, Partial Fourier, TWIST, oversampling, and matrix size for both nominal temporal resolution and impulse response characteristics. We found there to be a optimal simulation setting for balancing temporal accuracy of lesion enhancement with the measured amount of enhancement. Under conditions of only modifying TWIST from clinical settings, our findings suggest that using a pA of .22% and a pB of .25 provides an ideal combination of the two, improving the ability resolve the enhancement of a lesion within the breast compared to clinical settings.
Access status: Embargo ,
Diffusion Model Based Sampling with Metropolis-Hastings Correction
(2026) Hu, Yuang
This thesis investigates unbiased sampling from complex and higher dimensional prob-ability distributions using score-based diffusion models. While diffusion samplers have demonstrated strong generative performance, they inherently produce biased samples due to discretization errors and imperfect score estimation. To address this limitation, we de- velop a Metropolis–Hastings corrected diffusion framework that preserves the flexibility of diffusion-based proposals while guaranteeing asymptotic exactness. By deriving closed- form forward and reverse diffusion kernels, our method enables valid acceptance–rejection steps without approximations. Numerical experiments on multimodal synthetic targets demonstrate that our approach reduces sampling bias and improves posterior accuracy, while maintaining efficient mixing compared to traditional Markov chain methods. This work contributes to the fields of Bayesian statistics and machine learning by bridging deep generative diffusion models with theoretically rigorous MCMC, offering a practical tool for exact inference in challenging high-dimensional settings.
Access status: Embargo ,
KnotVision: Three-Dimensional Hand Motion Reconstruction and Quantitative Skill Assessment for Surgical Knot-Tying
(2026) Reid, Cameron Martine
Assessment of open surgical knot-tying skill remains largely subjective, time-intensive,and dependent on expert observation, limiting the scalability and consistency of technical- skills evaluation in surgical training. This thesis develops a markerless dual-view video framework to quantify knot-tying performance from hand-motion data. Using a benchtop knot-tying task recorded with two camera views, the proposed pipeline performs hand land- mark extraction, temporal synchronization, three-dimensional triangulation, and knot-level segmentation to reconstruct hand kinematics. Translational and rotational motion metrics are then computed to characterize timing, movement behavior, and dexterity during task execution. The results demonstrate that markerless hand tracking can provide objective, process-based measures of surgical knot-tying performance in a controlled setting. These findings support the feasibility of scalable video-based surgical skill assessment and establish a foundation for future hand-centered applications in surgical education and robotics.
Access status: Embargo ,
Associations Between Selected Maternal Empowerment Domains and Routine Infant Vaccine-Series Completion in Cambodia
(2026) Ding, Wenqing
Background: Childhood vaccination is a repeated behavioral process that requires caregivers to return for multiple scheduled visits. In settings where vaccines are provided free of charge and overall coverage is relatively high, remaining gaps may be more closely related to successful completion of repeated visits than to initial access alone. Different domains of maternal empowerment may therefore matter differently for routine infant vaccine-series completion.Objective: This study examined the association between selected maternal empowerment domains and routine infant vaccine-series completion in Cambodia. Methods: Data were drawn from the 2021–2022 Cambodia Demographic and Health Survey (CDHS). The analytical sample included 1,677 mother–child pairs with children aged 12–23 months. Maternal empowerment was assessed using two selected domains: economic agency and knowledge/information-related empowerment. The outcomes were completion of the three-dose DTP, OPV, and PCV infant vaccine series, as well as completion of all three infant vaccine series combined. In descriptive analysis, vaccination status was presented as zero-dose, incomplete, and complete. In regression analysis, due to small counts of zero-dose, zero-dose and incomplete were combined into the not-completed category. Binary logistic regression models were fitted in unadjusted and adjusted analyses, controlling for child characteristics, maternal background characteristics, and household socioeconomic and contextual factors. Results: In the analytical sample, the knowledge and information empowerment domain was more concentrated in the medium and high categories than the economic agency domain. Specifically, 18.17% of mothers were classified as low, 39.13% as medium, and 42.70% as high in the knowledge and information domain, compared with 25.85%, 46.89%, and 27.25%, respectively, for economic agency. Across the antigen-specific outcomes, completion was the most common vaccination status, although zero-dose and incomplete vaccination remained present. For OPV3, 7.35% of children were classified as zero-dose, 8.78% as incomplete, and 83.87% as complete; for DTP3, the corresponding proportions were 10.63%, 8.18%, and 81.18%; and for PCV3, they were 10.69%, 8.60%, and 80.70%. For the combined three-vaccine series outcome, 7.23% of children were zero-dose, 15.17% were incomplete, and 77.60% were complete. The most consistent pattern was observed for knowledge and information empowerment. Compared with children of mothers in the low group, those in the medium and high knowledge/information groups had significantly higher odds of completing OPV3, DTP3, PCV3, and all three infant vaccine series in adjusted analyses: OPV3 (medium: OR = 2.45, 95% CI: 1.46–4.12; high: OR = 3.53, 95% CI: 1.84–6.77), DTP3 (medium: OR = 2.29, 95% CI: 1.41–3.73; high: OR = 3.20, 95% CI: 1.76–5.84), PCV3 (medium: OR = 2.09, 95% CI: 1.30–3.36; high: OR = 2.95, 95% CI: 1.63–5.34), and the combined three-vaccine series outcome (medium: OR = 1.93, 95% CI: 1.24–3.02; high: OR = 2.60, 95% CI: 1.49–4.52). By contrast, economic agency showed a generally positive direction, but its associations were weaker and were not statistically significant in adjusted analyses. Conclusions: Among the retained maternal empowerment domains, knowledge and information empowerment was most consistently associated with routine infant vaccine-series completion in Cambodia. These findings suggest that, even in a context where routine childhood vaccines are provided free of charge, successful completion of repeated infant vaccination visits may also depend on mothers’ ability to access, understand, and use immunization-related information in ways that support continued use of vaccination services. Efforts to improve childhood vaccination outcomes in Cambodia may therefore benefit from strengthening communication, reminders, and follow-up support for repeated vaccination visits.
Access status: Embargo ,
Rethinking Dropout in Transformer-Based Neural Operators: Mechanisms, Effects, and Design Principles
(2026) Shao, Yuyuan
Transformer models have achieved significant success in natural language processing and computer vision and are now increasingly applied in scientific machine learning to solve complex partial differential equations. However, conventional training techniques, including dropout regularization, are frequently adopted from related fields without a critical evaluation of their appropriateness for partial differential equation (PDE) operator learning. This thesis investigates the effectiveness of dropout in transformer-based operator learning for elliptic PDEs through systematic experiments. Additionally, we introduce a correlated dropout framework to investigate the influence of spatial-frequency characteristics of dropout masks on results, comparing independent, low-frequency, high-frequency, and local Gaussian correlated dropout methods. Our findings consistently show that removing dropout improves accuracy in transformer-based elliptic PDE operator learning. Based on these results, we suggest removing dropout as a practical guideline for designing transformer-based operator learning models.
Access status: Open Access ,
Bayesian Joint Longitudinal-Survival Modeling with Time-Dependent Random Effects and Discontinuous Risk Intervals
(2026) Solarz, Kaitlyn Grace
Chronic pain is a common and persistent concern among cancer survivors, particularly among women with a current or prior incidence of breast cancer, and may influence patterns of opioid use and related adverse outcomes. Despite this clinical reality, existing approaches to modeling opioid-related risk typically rely only on baseline covariates and fail to account for the dynamic evolution of pain over time. We develop a Bayesian joint longitudinal–survival modeling framework to assess the relationship between chronic pain trajectories and time to opioid escalation using electronic health record data from a cohort of breast cancer survivors. The model links a longitudinal pain process with a time-to-event outcome through subject-specific random effects, treating pain as an internal time-varying covariate. We further incorporate clinically informed intervention windows (e.g., postoperative periods) during which subjects are not considered at risk for escalation. Posterior inference is performed via a Gibbs-within-Metropolis–Hastings Markov Chain Monte Carlo algorithm. Application of the proposed model to the study cohort data suggests that higher underlying pain burden is associated with increased risk of opioid escalation, while covariate effects reveal heterogeneity in risk across patient subgroups. These results demonstrate the utility of joint modeling approaches and underscore the importance of incorporating longitudinal pain dynamics into event risk modeling in this setting.
Access status: Open Access ,
Mapping the Local State: Operationalizing Industrial Strategy through Policy Portfolios
(2026) Liu , Yuejia
The political economy literature on China has long predicted that distinct local institutionallegacies should produce divergent models of development. However, empirically assessing these predictions has been hampered by a measurement gap: scholars often rely on simplistic proxies that treat the "black box" of policy design as opaque. This paper bridges this gap by simultaneously advancing policy measurement and unpacking the structural mechanisms of contemporary state intervention. First, using a novel LLM-assisted textual analysis of 1,123 provincial New Energy Vehicle (NEV) policy documents, I construct a high-resolution dataset of local policy portfolios. Second, I leverage this dataset to test the micro-behavioral logic driving local policy design. My analysis reveals a pattern of capacity-constrained convergence: while uniform central mandates drive rapid standardization in basic administrative tools, uneven local capacity drives stratification in high-complexity domains like R&D. Interrupted time series and network analyses demonstrate that this two-tier dynamic is governed by asymmetric political risk aversion and resource crowding-out, often resulting in symbolic compliance when mandates exceed local capabilities. Ultimately, this research reveals that contemporary local state agency has shifted from unrestrained fragmentation to tactically bounded diversity.
Access status: Open Access ,
UN Peacekeeping Stance and Civil Conflict Duration
(2026) Pan, Tianxiang
This study investigates how the United Nations' (UN) "stance" in civil conflicts affects war duration. Moving beyond the common assumption of institutional neutrality, this research introduces a key conceptual distinction: while traditional "bias" is often viewed as an a priori favoritism rooted in external interests, "stance" represents a purposive, reactive, and asymmetric pressure intentionally imposed by the UN as an independent political agent after observing battlefield behavior. Rather than focusing on the subjective perceptions of local actors, this research analyzes "stance" as a conscious strategic choice manifested in UN Security Council resolutions. The paper argues that neutral peacekeeping is more effective at shortening conflicts than any non-neutral intervention. The proposed mechanism is that neutrality facilitates conflict termination by allowing the UN to serve as a credible informational and relational mediator. In contrast, the imposition of asymmetric pressure, regardless of its normative motivation, tends to disrupt strategic balances and impede communication between belligerents. Using a Cox proportional hazards model on 210 civil conflicts (1946-2015), the empirical results strongly support this hypothesis: neutral UN involvement is significantly associated with shorter conflict duration. Conversely, the intentional adoption of a "stance" that is inclined toward the state or inclined toward rebels generally tends to prolong the conflict.
Access status: Embargo ,
Cross-Contrast Diffusion: A Synergistic Approach for Simultaneous Multi-Contrast MRI Super-Resolution
(2026) Wu, Yulu
Purpose: Diffusion-based deep-learning frameworks have been recently used in MRI resolution enhancement, or super-resolution. Multi-contrast MRI share common anatomical structures while holding complementary soft-tissue information. This study aims to incorporate both shared and contrast-specific features of multi-contrast MRI into a diffusion-based deep-learning framework for simultaneous multi-contrast MRI super-resolution.Methods: Public IXI brain dataset consisting of 576 healthy participants was used. High resolution (HR) multi-contrast MRI (T1-w and T2-w, 0.9375mm x 0.9375mm in 2D) were defined as groundtruth. Low resolution (LR) multi-contrast MRI (T1-w and T2-w, 3.75mm x 3.75mm in 2D) were simulated by low-pass filtering in K-space and used as input. LR T1-w and T2-w images were combined into a dual-channel representation in the forward diffusion process. In the backward process, a cross-contrast encoder with a spatial self-attention module was added to capture shared structural features and contrast-specific information. The separation was guided by a disentanglement term. All extracted features were processed through a Squeeze-and-Excitation module for channel-wise weighting. Charbonnier loss and a progressive training strategy were employed for training stability. The method was examined on the IXI dataset with training, validation, and testing split of 500:6:70 and compared with several state-of-the-art super-resolution methods, including Bicubic Interpolation, EDSR, SwinIR, Guided Diffusion, MINet, MASA-SR under the metrics of PSNR and SSIM.Results: The method achieved simultaneous high-quality T1-w and T2-w super-resolution, enhancing resolution by 2 and 4 times. On 2x super resolution, the PSNR/SSIM were 37.42 dB/0.9869 for T1-w and 37.70 dB/0.9879 for T2-w MRI. On 4x super resolution, the PSNR/SSIM were 31.61 dB/0.9570 for T1-w and 30.85 dB/0.9443 for T2-w MRI. Notably different from other multi-contrast methods, our method does not need HR MRI as input, while achieving the highest PSNR and SSIM on both magnification factors among the state-of-the-art (SOTA) methods. Conclusions: To our knowledge, this is the first study of simultaneous multi-contrast MRI super-resolution using diffusion-based framework. Our network effectively achieves high quality super-resolution of two MRI contrasts without need of high-resolution guidance. Future work includes extension studies in additional contrasts and 3D image scenarios.
Access status: Embargo ,
Federated Learning for CT-Based Lung Cancer Subtype Classification: A Multi-Dimensional Methodological Comparison
(2026) Yuan, Yue
\abstract{\textbf{Background:}} Histological subtype classification of lung cancer is clinically important because the major subtypes differ in biological behavior, diagnostic pathways, treatment choice, and prognosis. Computed tomography (CT) is routinely acquired for lung cancer patients to capture tumor-related phenotypic information. CT-based deep learning methods have shown promise in differentiating lung cancer subtypes by data driven learning. Multicenter training offers a favorable route to more accurate and robust models, while direct data sharing across institutions is often limited by privacy concerns and data-governance restrictions. Federated learning (FL) is a privacy-preserving distributed learning mechanism for multicenter data utilization.\textbf{Purpose:} This study aimed to compare centralized learning (CL) and federated learning for four-class CT-based lung cancer subtype classification, with particular emphasis on examining multiple dimensions of condition, including input image size, prediction strategy, neural network backbone, and data heterogeneity.\textbf{Materials and Methods:} The dataset was derived from the public Lung-PET-CT-Dx collection released through The Cancer Imaging Archive. After modality filtering and preprocessing, the final analysis set consisted of 820 image-level CT acquisitions, including 606 adenocarcinoma, 80 small cell carcinoma, 10 large cell carcinoma, and 124 squamous cell carcinoma images. Four pretrained deep learning backbones, namely ResNet-50, ConvNeXt-Tiny, Swin-Tiny, and MambaOut-Tiny, were evaluated within a unified five-fold stratified cross-validation framework. Experiments were conducted under two training paradigms, centralized learning and federated learning, two decision strategies, single-model and soft-voting ensemble prediction, two input representations, full-slice and ROI-only. The optimal condition for federated learning under homogenous data conditions was transferred to the examination of heterogeneous data conditions. Heterogeneity in dataset size, image feature, and a mixture of the two were simulated and explored. Federated learning followed a FedAvg-style workflow with three simulated clients. Performance was assessed using macro-F1 as the primary model-selection metric, together with accuracy, balanced accuracy, macro-AUC, sensitivity, specificity, and class-wise metrics.\textbf{Results:} Under homogenous data condition, the strongest federated result achieved a macro-F1 of $0.8337 \pm 0.0299$ and a macro-AUC of $0.9365 \pm 0.0107$ (single-model, full-slice), was close to the strongest centralized result, which achieved a macro-F1 of $0.8450 \pm 0.0237$ and a macro-AUC of $0.9372 \pm 0.0119$ (single-model, full-slice). Ensemble prediction did not surpass the best individual backbone in macro-F1, but the highest macro-AUC in the study was achieved by the centralized full-slice ACC-weighted ensemble ($0.9687 \pm 0.0134$). Full-slice input consistently outperformed ROI-only input across both centralized and federated settings. Averaged across backbones, full-slice input increased macro-F1 from $0.5435$ to $0.7713$ under centralized learning and from $0.5138$ to $0.7726$ under federated learning. Under several simulated heterogeneous data conditions, federated learning showed in general lower average macro-F1 than centralized learning.\textbf{Conclusion:} Under homogenous data conditions, FedAvg federated learning maintained competitive classification performance for CT-based lung cancer four-type classification without requiring raw data sharing. Ensemble learning did not improve macro-F1 beyond the best single model, while enhanced probability-based discrimination, particularly macro-AUC. Full-slice CT input consistently outperforms tightly cropped ROI-only input, suggesting that contextual image information may play an important role in multiclass lung cancer subtype classification. Under simulated heterogeneous conditions, FedAvg federated learning showed lower average macro-F1 than centralized learning. Future studies include more comprehensive study of heterogeneous conditions and study of alternative federated learning strategies to reduce the performance gap between FL and CL.
Access status: Open Access ,
Ideology, Elite Cohesion, and Authoritarian State Building: Evidence in Southern Song China
(2026) Xiao, Xiongfu
How do authoritarian states foster elite cohesion and loyalty, and in turn, state capacity? Existing scholarship emphasizes material co-optation or war preparation, but often overlooks the role of ideology as an endogenous mechanism of state building. This paper develops a formal model and applies network dependency theory to compare purges and ideology as alternative strategies of elite integration. Using original datasets of 1,488 elites coded from primary historical sources with reference to HBDB dataset across three historical episodes of the Song dynasty, I conduct Social Network Analysis and regression analysis to evaluate the mechanisms' effects. The results show that purges fail to enhance elite cohesion, while the institutionalization of Neo-Confucianism significantly strengthened elite networks and increased wartime loyalty. The study contributes to Historical Political Economy by demonstrating that durable state capacity depends not only on coercion and institutions, but also on the symbolic infrastructures of ideology.
Access status: Embargo ,
Analysis of PET Image Quality Using 2D vs 3D ROIs in Small Spheres
(2026) Simmons, Justin
Modern positron emission tomography (PET) systems have substantially improved insensitivity, timing performance, and reconstruction methods; however, the conventional NEMA image-quality phantom remains limited in evaluating sub-centimeter lesion de- tectability. This work investigated the image quality of small-sphere PET using an exper- imental image-quality phantom designed to extend conventional NEMA-style assessment into the sub-centimeter regime, along with an automated analysis workflow for reproducible metric extraction. A second whole-body phantom containing multiple small spheres was also evaluated in a more anthropomorphic geometry. An automated MATLAB-based pipeline was developed to process PET DICOM data, localize sphere centers, define regions of interest, and compute contrast recovery coefficient (CRC), coefficient of variation (CoV), and contrast-to-noise ratio (CNR). The experimen- tal phantom was evaluated across multiple PET/CT systems and reconstruction settings. Results showed that averaged-frame localization closely matched static acquisition, with a mean whole-phantom percent error of less than 1%. CRC and CNR increased with sphere diameter, and the greatest reconstruction-dependent differences were observed for the small- est spheres. In the whole-body phantom, the expected tradeoff between contrast recovery and background variability was observed. Overall, this work demonstrates the value of smaller-sphere phantom designs for more sensitive evaluation of modern PET image quality.
Access status: Embargo ,
Synergizing Convolutional Backbones with Inductive kNN-Residual Graph Neural Networks for Preoperative Diagnosis of MTM-HCC
(2026) Hao, Yifei
AbstractPurpose: Macrotrabecular-massive hepatocellular carcinoma (MTM-HCC) is an aggressive subtype of liver cancer with limited imaging signatures, making accurate noninvasive diagnosis challenging. This study introduces a novel residual network 101 based k-nearest neighbors (kNN) graph assisted Residual Graph Convolution Network (ResNet101-kNN-ResGCN) designed for the preoperative classification of macrotrabecular-massive hepatocellular carcinoma (MTM-HCC) using multicenter CT data, with the goal of investigating whether leveraging local structural relationships among image features could improve diagnostic accuracy. Materials and Methods: This retrospective study included 503 patients from two institutions, comprising an internal cohort of 372 patients (87 MTM-HCC and 285 non-MTM-HCC) and an external cohort of 131 patients (34 MTM-HCC and 97 non-MTM-HCC). Two-dimensional slices from portal venous phase contrast-enhanced CT scans were used. Images were manually segmented by experienced radiologists to define regions of interest. The internal cohort was split into training and validation sets of 85% and 15%, with the external cohort held as an independent test set. An ResNet101-kNN-ResGCN architecture that combines convolutional feature extraction with graph-based relational modeling was designed. Deep feature are extracted using a ResNet101 backbone, followed by the dynamic construction of an inductive kNN graph within each batch in the feature space. This batch-wise graph encodes local feature similarity among samples within each mini-batch and serves as a structural prior for subsequent residual GCN layers. It should be noted that this graph reflects local relationships within the batch rather than global relationships across the entire dataset. Two residual GCN blocks based on GCN progressively transform the backbone features into representation, which is finally fed into a fully connected layer for binary classification. A class-weighted cross-entropy loss to mitigate class imbalance. The area under the receiver operating characteristic curve (AUC), Accuracy (ACC), sensitivity (SEN) and specificity (SPE), and were used to evaluate diagnostic performance. Results: The ResNet101-kNN-ResGCN architecture demonstrated improved diagnostic performance compared to baseline and alternative GCN configurations. The model achieved a peak AUC of 0.867 (95% CI: 0.803-0.919). In contrast, the standalone ResNet101 baseline yielded a lower AUC of 0.653 (95% CI: 0.544-0.765), while the chain-structured GCN and ResGCN reached an AUC of 0.815 (95% CI: 0.738-0.886) and 0.836 (95% CI: 0.762-0.897), respectively. Conclusions: This study demonstrates that a kNN-based residual GCN can capture discriminative patterns in the feature space associated with MTM-HCC from CT scans, offering a new approach for the deep learning based non-invasive preoperative risk stratification in clinical practice. The results indicate that incorporating residual connections within a kNN-based graph convolution framework may provide performance gain over both sequential GCN and standalone ResNet101 structures.
Access status: Open Access ,
The Organization of Perception: Narrative, Technology, and Imperial Fantasy in Japanese Sci-Fi Anime
(2026) Li, Jiaying
Science fiction (SF) anime flourished in Japan during the 1970s and 1980s. This thesis discusses why this specific genre flourished in the era when grand narratives shrank and the reality was reconstructed. It focuses on two case studies: Mobile Suit Gundam (1979) and Megazone 23 (1985). Drawing on close reading of their narrative structures and visual language,and situating them within the wider transmedia forms through which they circulated, the thesis examines how these works both reflected and critically engaged with the social conditions of their time. It argues that SF anime emerged as a cultural form capable of reorganizing fragmented time, mediating technological perception, and exposing the constructed nature of everyday reality.
Access status: Embargo ,
Data-Driven Synthesis of Spiculated Breast Phantoms: A User-Interactive Framework for Modeling and Clinic Image Integration
(2026) Wan, Wenbo
The advancement of virtual clinical trials (VCT) and artificial intelligence (AI) systems in breast imaging requires large, diverse datasets with reliable annotations. However, the limited availability of clinical datasets and the uncontrollable variability of biological lesion characteristics create a critical bottleneck. While computational phantoms offer an alternative, current methodologies often lack clinically realistic morphological irregularity and fail to provide a physics-aware image fusion technique, resulting in unnatural artifacts when inserted into clinical host images.This thesis proposes a novel, multi-stage interactive software framework for the procedural generation of 3D spiculated breast mass phantoms and their physics-aware insertion into full-field digital mammograms (FFDM). The framework utilizes clinical FFDMs to extract a 2D mass contour, generating a 3D core via a modulated extrusion technique adapted by local intensity and parabolic shape factors. Complex spiculation patterns are simulated using an iterative fractal branching algorithm analogous to a stochastic Lindenmayer system (L-system). Finally, the 2D projection of the synthesized phantom is integrated into clinical images using a physics-based additive superposition model with a targeted brightness calibration factor, ensuring rigorous contrast preservation across different breast tissue densities. Reader studies and quantitative radiomics analysis (evaluating circularity, solidity, and Weber Contrast) demonstrated that the generated phantoms achieve moderate-to-high visual realism. The physics-based insertion algorithm successfully calculated realistic attenuation levels regardless of background parenchymal complexity. While the framework effectively replicates benign lesion features, future refinement through radiomics-guided modeling is needed to better capture the highly chaotic morphology of high-grade malignancies. Ultimately, this tool provides a practical, cost-effective solution for systematically evaluating breast imaging technologies.
Access status: Embargo ,
Fiction, Friction: A Sengoha Encounter with Lotus
(2026) Cai, Jiayang
This thesis examines Japanese participation in the Afro-Asian Writers’ Association (AAWA) and its journal Lotus: Afro-Asian Writing (1968–1991), recovering a neglected history of Japan’s engagement with Third World internationalism. While recent scholarship has treated Lotus as a site of transnational solidarity and contradiction, Japanese contributions have remained largely absent from these accounts.The thesis argues that Japanese participation was structured by a grammar of victimhood and resistance first articulated at the 1958 Tashkent conference, which positioned the Japanese as victims and resisters of imperialism. This framing enabled inclusion in an anti-colonial project despite Japan’s own imperial past, but constrained how that past could be represented. The result was a persistent disjunction between critical prose, which gradually acknowledged Japanese complicity, and literary selections, which remained confined to narratives of victimhood and anti-war consciousness. Through close readings of works by Ayukawa Nobuo, Kijima Hajime, and Hotta Yoshie alongside texts by Palestinian authors in Lotus Issue 4 (1970), the thesis proposes that the journal generates a form of contrapuntal awareness in which the limits of Japanese anti-imperial literature become legible only in relation to writing from the colonized world — an effect produced through editorial juxtaposition rather than explicit critique, and all the more revealing for being unacknowledged.
Access status: Embargo ,
Respiration-Related Artifacts in PET/CT Imaging
(2026) Moeser, Katherine
Discrepancy between the diaphragm position in PET and CT during PET/CT studies can lead to artifacts in the PET images due to improper attenuation correction. This artifact is an artificially depleted region at the liver dome that can potentially obscure lesions. We are developing a framework for quantifying the degree of mismatch and severity of artifacts on patient images and have performed phantom studies to illustrate the effect. A phantom was built with semi-cylinder lungs and 2 cm spheres below the lungs. The phantom was scanned with no motion, and then with incremental 1 cm shifts up to 5 cm in the axial direction to achieve PET/CT mismatch and blurring of PET. The spheres were filled with 6:1 and 3:1 radioactivity concentrations compared to the background. PET images were acquired and reconstructed using two different scanners with (AC) and without (NAC) attenuation correction and with and without time of flight (TOF) information. Additionally, patient images from FDG and PSMA studies were anonymized and evaluated for the presence of the artifact and the difference in diaphragm position. The location of the apex of the diaphragm was determined in each image. These patients were instructed to hold breath at end tidal volume. NAC PET images were used since the AC images are affected by the CT. A total of 168 FDG studies and 84 PSMA studies were evaluated.
Access status: Open Access ,
Ultra-Processed Food Consumption Among Infants at 12 Months of Age: Prevalence and Contributions to Energy, Sodium, Added Sugars, and Saturated Fats
(2026) Ma, Boyi
Ultra-processed foods (UPFs) have increased substantially in early childhood diets, yet limited research has examined the extent to which UPFs contribute to energy and nutrient intake among infants during the complementary feeding period. This study aimed to characterize consumption of ultra-processed foods among infants at 12 months of age and to quantify the contribution of UPFs to total daily intake of energy, sodium, added sugars, and saturated fats.Dietary intake data were obtained from a sample of 237 infants at 12 months of age. Infant dietary intake was assessed using a single 24-hour dietary recall and analyzed with the Nutrition Data System for Research (NDSR). Foods and beverages were classified according to the NOVA food classification system. Descriptive analyses were conducted to estimate the prevalence of UPF consumption and to calculate the mean daily intake of energy and selected nutrients from infant formula, other ultra-processed foods, and non-ultra-processed foods. All infants in the sample consumed at least one ultra-processed food when infant formula was included in the UPF classification, and 94.1% consumed ultra-processed foods excluding formula. Ultra-processed foods contributed 61.4% of total daily energy intake, 52.1% of total sodium intake, 98.5% of total added sugars intake, and 56.6% of total saturated fats intake. Infant formula was a major contributor to total energy, added sugars, and saturated fats intake, while other ultra-processed foods contributed substantially to sodium intake. These findings indicate that ultra-processed foods constitute a substantial proportion of infants' dietary intake at 12 months of age. The results highlight early exposure to ultra-processed foods and their contribution to key nutrient concerns during a critical period of dietary development.
Access status: Open Access ,
ConnGA: Genetic Algorithm-Based Framework for Optimizing DTI Tractography in Pre-Clinical Murine Models of Alzheimer’s Disease
(2026) Canamedi, Varun
We have established methods to acquire 25 µm resolution Diffusion Tensor Imaging (DTI) in postmortem mouse brain models of Alzheimer’s disease (AD). However, the utility of these advances remains limited by suboptimal post-processing pipelines. Here, we develop a genetic algorithm (GA) to identify tractography parameters required to enhance disease-specific disruptions in brain structural connectivity in the BXD-77 murine model. Given the susceptibility of tractography to false positives, we evaluate the generalizability of GA-optimized tracking across multiple BXD strains (BXD-101, BXD-65, BXD77, and BXD-32) and assess how variations in tractography parameters influence the relationship between amyloid pathology and structural connectivity. We also quantify the relative importance of individual tractography parameters in determining group separation. GA-optimized tracking produced substantial gains in group separability and showed consistency with phenotypic gradients observed across multiple strains, while demonstrating improved agreement with histological data. This work presents a robust optimization framework that enhances the reliability and generalizability of tractography-based connectomic analyses in preclinical models of neurodegeneration.
Access status: Embargo ,
Computationally Designed Enzyme Inhibitors Enable Rewiring of Metabolic Networks
(2026) Kim, Yeonseo
Traditional metabolic engineering has historically relied on static genetic interventions, such as gene knockouts or constitutive overexpression, to redirect carbon flux. However, these approaches are fundamentally limited by the physiological conflict between cellular growth and product synthesis. While dynamic control strategies like CRISPR interference and targeted proteolysis have been developed to decouple these phases, they primarily operate at the gene level by modulating enzyme abundance. Such coarse regulation can lead to physiological instability when entire enzymes are removed and often relies on host-specific regulatory machinery, limiting portability across different organisms.To address these limitations, this thesis presents a new modality for metabolic control based on computationally designed protein binders that modulate enzyme activity directly through molecular recognition. Glucose 1-dehydrogenase (GDH) was selected as a model system for this proof-of-concept study due to its central role in carbon partitioning within the gluconate-bypass (GBP) strain. Candidate nanobody (VHH) binders were generated using the RFantibody generative design framework, targeting three functionally distinct epitopes: the substrate-binding site (Site A), the catalytic active site (Site B), and the NADP+ binding site (Site C). A multi-stage in silico screening pipeline was implemented to enrich the design pool, utilizing Molecular Mechanics / Generalized Born Surface Area (MM/GBSA) for energetic ranking followed by molecular dynamics (MD) simulations to assess the structural stability and interface integrity of the predicted antibody-enzyme complexes. The results of this study demonstrate that de novo binders can be rationally designed to interact with key functional regions of an enzyme and potentially alter its catalytic activity. Computational analysis revealed that designs targeting Sites A and B achieved more favorable energetic distributions and structural equilibration compared to Site C, likely due to the more complex surface features surrounding those pockets. This work establishes the feasibility of using computational protein design to create post-translational regulatory tools that decouple enzyme abundance from metabolic flux. By providing a modular, portable, and tunable alternative to traditional genetic regulation, this framework offers a promising strategy for the sophisticated rewiring of metabolic networks.

Browse

Recent Submissions