Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
PET and SPET: A Two-Stage Pipeline for Pulmonary Embolism Detection from CTPA Scans
No full textPulmonary embolism (PE) is a life-threatening condition with a high mortality rate of 30%. Immediate treatment is crucial to improve treatment outcomes, but PE diagnosis often takes multiple days due to the limited availability of radiologists to analyze computed tomography pulmonary angiography (CTPA) images, which involves taking multiple X-ray images (often 100s of slices) of the chest region. Therefore, automation of diagnosis of PE can significantly improve patient outcomes. We use the RadFusion dataset to improve upon prior work to diagnose PE using the CTPA scans as well as using electronic health records. The labeling method and pre-processing of EHR data was improved, with over 50% of the features being identified as redundant. Correlation based analysis was done to select key EHR features, drastically reducing the number of features needed. A novel two stage pipeline was developed for analyzing CTPE images. This approach consists of one model, PE-Transformer (PET), to analyze a window of contiguous CTPE slices (Chunks), followed by another model, Sequential PE-Transformer (S-PET) to aggregate information across multiple chunks. The PET model architecture improved AUROC by 1.8% due to using a standard DinoV2 + Transformer architecture rather than a custom architecture. The S-PET model improved the accuracy by an additional 1.2% due to aggregating information across chunks. Distilling the metadata features to just the top 16 most important features and using a random forest classifier resulted in the highest AUROC of 0.79 compared to the 0.76 when using all the metadata features. The analysis supports the results from larger datasets such as INSPECT and bridge the gap in prior work (RadFusion) which indicated that EHR data was more accurate than PE data. We also demonstrate that modern self-supervised backbones trained on web scale data offer superior performance, reducing the need for custom architectures. This research also shows that very few EHR features contribute to accuracy, reducing the need for collection of large amounts of EHR data. The current approach isn't end-to-end trainable and separates chunk-level and patient-level models; in the future, we aim to explore unified models and more complex backbones. The code and checkpoints are available at: https://github.com/Anisha234/ASSIP_research_pene
Investigating autophagy and mitophagy in 12Z endometriosis cells via exposure to pharmacological inhibitors
No full textEndometriosis is a chronic condition in which endometrial tissue spreads outside the uterus and causes inflammation. As these cells enter oxidative stress conditions, they rely on autophagy and mitophagy–cellular recycling mechanisms–to survive. Thwarting these processes may prohibit further growth. This experiment applied medications suspected to inhibit autophagy and mitophagy to endometriosis cells to study the effects on growth and protein pathways. Endometriosis 12Z cells were cultured and divided into two assays. In the first, cells were treated with chloroquine phosphate, metformin, lidocaine, or combinations of these for different time stamps. After treatment, cells were lysed and analyzed using reverse phase protein array (RPPA). In the second experiment, endometriosis cells were cultured and scratched to assess migration and growth response to metformin, lidocaine, tamoxifen, and chloroquine phosphate. Following treatment, cells were lysed and analyzed using western blot, RPPA, and immunohistochemistry. Morphological analysis of 12Z cells in the first experiment revealed responses to chloroquine phosphate, including rod-shaped distortion, refractility, and large cytoplasmic vacuoles. These effects were amplified by co-treatment with lidocaine or metformin, leading to pronounced cell swelling and vacuolization—signs of disrupted autophagy. Preliminary data suggests that 12Z cells exposed to drugs, such as chloroquine phosphate, left wider gaps in scratch assays, suggesting slower growth and increased stress responses. Metformin+Lidocaine increased PHGDH, Calbindin, and AMPK alpha Thr172 in the 48-hour time course. Chloroquine phosphate increased proteins in glucose, autophagy, and oxidative stress pathways. Chloroquine-treated cells induced autophagy-disruption and stress, while Metformin+Lidocaine elevated the levels of autophagy pathway proteins. Ongoing research aims to investigate the pathways involved
Integrating Gene and Clinical Data to Overcome Melanoma Cancer with Pembrolizuamb
No full textMelanoma accounts for only about 1% of all skin cancer cases, but it causes the majority of skin cancer deaths. It is especially dangerous for teens and young adults, with about 1 in 50 Americans diagnosed in their lifetime. This project used machine learning to explore whether gene expression data could help predict how melanoma patients respond to pembrolizumab, an immune-based cancer treatment that targets the PD-1 receptor. Public datasets including GSE91061 were used, which include samples grouped by treatment outcomes such as complete or partial response, stable disease, or disease progression. The data was carefully processed through background correction, normalization, and filtering, followed by differential gene expression analysis using the limma linear modeling method. Ten-fold cross-validation was also used to prepare the dataset for predictive modeling. While the results revealed patterns in gene activity linked to treatment outcomes, further research with expanded datasets is needed to improve the reliability of future predictions
Integrating Gene and Clinical Data to Overcome Chemoresistance in Colorectal Cancer
No full textColorectal cancer remains a major health challenge due to frequent drug resistance. Current treatment plans vary widely, partly because tumors can behave differently based on the genetic alterations present in the tumor. The inconsistency makes it harder to find the best treatment for individuals. Common treatments usually involve chemotherapy drug combinations such as FOLFOX (5-fluorouracil, leucovorin, and oxaliplatin) and FOLFIRI (5-fluorouracil, leucovorin, and irinotecan). Here, we enhance a gene expression panel that predicts how colorectal cancer patients will respond to drugs by stratifying gene signatures for samples from patients receiving different drug combination treatments and those with varying tumor locations. The model demonstrates strong predictive accuracy in categorizing patients into groups based on their likelihood of responding (or not) to various chemotherapy drugs, highlighting how a patient's geographical location can influence the drugs they receive and the model's effectiveness in this context. The results of our work suggest that integrating gene expression profiling with real-world clinical factors can improve personalized treatment strategies, resulting in better outcomes by overcoming drug resistance in colorectal cancer
Healthcare Deserts in Virginia: Rural-Urban and Socioeconomic Disparities in Access to Care
No full textMany rural communities in the United States live in healthcare deserts, areas without access to care within a reasonable time, leading to higher rates of mortality and chronic disease compared to urban populations. While several studies examine healthcare access at an aggregate level, few focus on the unequal distribution of healthcare deserts across rural and urban areas and the related socio-economic disparities, especially in Virginia (VA). To address this gap, this study measures healthcare deserts in VA and explores their rural-urban and socio-economic drivers. We perform our analysis at both county and census tract levels, using different travel time thresholds and multiple provider types to understand how health desert measures vary across spatio-temporal scales and healthcare needs. Using hospital location data from the Dewey database and their NAICS codes, we classify hospitals into five categories: (1) physicians, (2) outpatient care and community health centers, (3) general medical and surgical hospitals, (4) specialty hospitals, and (5) mental health facilities. We also include socio-economic variables such as poverty rates, racial composition, vehicle ownership, and health insurance coverage, along with rural-urban classification data from the American Community Survey and the US Department of Agriculture. For each provider type, we calculate an access index representing the number of facilities reachable within 15- and 30-minute driving thresholds. Census tracts and counties with an access index of zero are classified as healthcare deserts. Next, we use binary logistic regression to examine how socio-economic and rural-urban factors influence the likelihood of a county or census tract being a healthcare desert. Our results reveal that between 12% and 50% of counties and 30% to 42% of tracts are considered health deserts across both time thresholds and all five provider types. Rural areas with additional socio-economic disadvantages are significantly more likely to be classified as health deserts than urban areas. These findings can help practitioners and policymakers identify the most healthcare-disadvantaged communities and develop targeted investments and policies to ensure equitable healthcare access throughout Virginia
The Architecture of Risk: A Cross-Sectional Analysis of Governance, Regulation, and Market Behavior in Cryptocurrency Exchanges
No full textThe proliferation of cryptocurrency exchanges is central to the digital asset economy, yet the sector is plagued by operational failures and regulatory enforcement actions. Existing risk assessments often rely on simplistic metrics like trading volume, lacking a systematic framework for evaluating the qualitative factors that drive an exchange's risk profile. This study addresses this gap by constructing a comprehensive, multi-dimensional database of over 250 centralized cryptocurrency exchanges. Through a systematic due-diligence process utilizing public records, regulatory filings, and legal documents, we mapped each exchange’s governance structure, regulatory frameworks, product offerings, and geographic exposures. Our analysis of this database highlights a strong correspondence between the permissiveness of an exchange's primary jurisdiction and its operational risk. Notably, a clear pattern emerges where exchanges operating in jurisdictions with weak regulatory oversight and opaque ownership structures are more likely to offer high-leverage derivative products and be implicated in enforcement actions. This research provides a foundational framework and a structured dataset for regulators and investors, enabling a more nuanced assessment of an exchange’s operational and legal structure and moving beyond surface-level metrics to a deeper understanding of counterparty risk
Unpacking Tokenized Global Bond Markets: A Due Diligence Analysis of RWA Products
No full textTraditional fixed-income markets rely on legacy infrastructure and intermediated settlement systems, limiting accessibility, transparency, and efficiency. In response, tokenization of real-world assets (RWAs), such as government and corporate bonds, has emerged as a promising mechanism for increasing programmability, enabling fractional ownership, and expanding global access. This study applies a standardized due diligence framework to examine 15 tokenized global bond products listed on rwa.xyz as of June 2025, analyzing offerings across key jurisdictions (France, Switzerland, Luxembourg, Mexico, the U.S., and the Cayman Islands) and protocols (Ethereum, Solana, Avalanche, Plume, and XRP Ledger). We combined legal document review with on-chain analytics (e.g., Etherscan, Arkham) to assess regulatory, operational, and technical dimensions. Our analysis of regulatory frameworks revealed significant variation: 3 of 15 products adopt regulated EU structures (e.g., the UCITS Directive), while 9 are domiciled in jurisdictions with crypto-specific legal regimes such as Switzerland's DLT Act and MiFID II-compliant structures. Despite small market capitalizations (median <$6M) and limited secondary market engagement, several products feature clear custody relationships, zero or minimal fees, and stable issuance practices. However, transparency around redemption rights and peg mechanisms remains inconsistent across offerings. In conclusion, while tokenized global bonds are in early stages, they reflect meaningful experimentation in bridging traditional finance with decentralized infrastructure. This work provides a foundational dataset for further regulatory, economic, and technical evaluation of tokenized debt markets
Evaluating Intel TDX as a Secure, Cost-Effective Solution for Replicating and Validating Academic Research
No full textAs the reproducibility of academic papers’ findings becomes a standard for credible scientific research, many journals now mandate authors to submit their code and data. However, replication packages often contain unvetted code, which, when executed in unprotected computing environments, can pose security and integrity risks to editorial offices and conference organizers responsible for verifying results. Additionally, it can risk leaking sensitive datasets or proprietary algorithms, which could expose confidential information. Current replication practices often lack hardware-based protection, leaving systems exposed to potential vulnerabilities during code execution. This study evaluates Intel’s Trust Domain Extensions (TDX), a confidential computing technology that runs code in secure, isolated environments, as a safer alternative for validating academic research. The evaluation tests replication packages from recently published Management Science papers in the finance, investment, accounting, and political science departments by deploying their code on TDX-enabled virtual machines (VM) hosted on commonly utilized cloud platforms such as Google Cloud and Microsoft Azure. Outputs generated in the TDX environment were compared to results reported in each paper to verify and evaluate reproducibility. Replication tasks were benchmarked on key metrics such as cost (measured in credits used) and runtime. On average, replication attempts across selected papers took 1.33 credits and ran for 2.22 hours, with longer run times associated with missing software dependencies, unavailable datasets, and proprietary platforms. These results imply that Intel TDX provides a practical and secure solution for executing academic replication packages at a reasonable cost and runtime, especially for papers with well-documented materials
Recurring Pattern Discovery on Time Series via Large Language Models
No full textTime Series Motif Discovery aims to uncover recurring patterns, or motifs, within large-scale time series data. These motifs have been shown to be valuable for a variety of downstream tasks, including forecasting, anomaly detection, and classification. Prior studies have demonstrated that inferring Context Free Grammars (CFGs) from discretized time series is an effective approach for discovering such patterns. CFGs capture structured patterns by defining rules that generate valid subsequences, similar to how Artificial Intelligence (AI) systems interpret human language. Traditional CFG inference algorithms like Sequitur adapt a bottom-up, greedy approach that identifies local repetitions early without considering larger, globally optimal patterns. In contrast, to address the limitations of existing approaches, we propose a top-down approach, inspired by human pattern recognition. Specifically, leveraging the Llama-3.1-8B-Instruct Large Language Model (LLM), we prioritize the discovery of high-level structures before decomposing into local, finer patterns. We develop a system to instruct the LLM to produce CFGs given a set of 1000 complex strings, each with 900 characters in length. The CFGs generated by the LLM, and subsequently the repeated patterns described by the CFGs, are evaluated against those from Sequitur using annotated ground truth. Preliminary experiments suggest that the LLM-generated CFGs are more compact compared to those from Sequitur. We anticipate that this approach will have significant impact in downstream applications including motif discovery and sequence compression, and ultimately more AI-driven analysis of time series data
Novel AI-Augmented Simulation Games for Water Security and Resilience
No full textWater scarcity is a growing global concern, particularly in agricultural regions where drought conditions threaten social stability. Non-governmental organizations (NGOs) and state-level water resource boards (WRBs) may offer monetary and non-monetary incentives to guide farmers toward socially beneficial and resilient water use. However, the complexity of interdependencies within agricultural water systems and human responses makes it difficult for NGOs and WRBs to optimize incentive schemes. We developed a simulation using the Godot game engine to reflect the roles NGOs and WRBs may play in water management during drought seasons. This single-player simulation models water distribution dynamics among three AI-controlled farms—upstream, midstream, and downstream—along a shared river in a drought-prone rural region. The player assumes the role of an NGO or a state WRB, offering seasonal monetary incentives over eight simulated seasons to influence farmer behavior under variable water availability. Each farmer AI agent operates based on unique attributes, including water permit volume, efficiency, stored water, robustness, welfare, and satisfaction thresholds. Farms independently decide whether to accept incentives, how much river water to use or store, and how to deploy stored water during short-term droughts. Key game mechanics, such as upstream-first water allocation and decentralized storage decisions, capture real-world challenges in achieving collective water resilience. Global variables such as supply, stored supply, efficiency, robustness, and social welfare allow comparison between player strategies and a game-theoretic baseline. Preliminary results suggest the system effectively illustrates trade-offs between individual decision-making and collective resilience. Ultimately, maintaining water resilience and social welfare in such systems requires not only effective incentive design but also a systems-level understanding of interdependent water use