18693 research outputs found
Sort by
SMARTER DISEASE DETECTION FROM ELECTRONIC HEALTH RECORD DATA: AN END-TO-END AI-AUGMENTED PIPELINE FOR COMPUTABLE PHENOTYPING
No full textElectronic Health Records (EHR) contain a wealth of structured and unstructured patient data that can be leveraged for computable phenotyping, the process of algorithmically identifying patient cohorts with specific diseases or conditions. Traditional rule-based phenotyping approaches, while interpretable, often struggle with scalability, portability across institutions, and effective use of unstructured clinical narratives. Recent advances in large language models (LLMs) present new opportunities for synthesizing complex free-text information into concise, clinically meaningful representations. However, integrating LLMs into phenotyping workflows requires careful design to maintain transparency, interpretability, and measurable uncertainty—features essential for clinical adoption and downstream applications such as decision support.
We developed an end-to-end, multimodal phenotyping pipeline that integrates structured EHR data with LLM-derived insights from unstructured clinical notes to improve disease classification. Using diabetes phenotyping as a proof-of-concept, the framework begins with a logistic-LASSO model trained on structured EHR features to generate patient-level predicted probabilities. Initially, augmentation targeted cases with intermediate probabilities—where uncertainty was highest and structured data alone was insufficient for accurate classification—by prompting an LLM to classify disease status from retrieved clinical notes. LLM-derived classifications were added to the structured predictor set as a three-level categorical variable indicating whether the patient was (1) not flagged for LLM augmentation, (2) LLM-classified as disease-absent, or (3) LLM-classified as disease-present. Compared with both a traditional rule-based phenotype and the structured-only logistic-LASSO, this probability-thresholding approach improved all measured performance metrics, demonstrating the added value of targeted unstructured data insights. Nonetheless, reliance on manually defined thresholds limited generalizability.
To address these limitations, we advanced to an ensemble-guided LLM-augmentation strategy. Here, a diverse set of base learners trained on structured data flagged cases for augmentation based on disagreement, eliminating subjective thresholds and offering an objective, adaptable selection criterion. This improved identification of patients most likely to benefit from LLM augmentation, and the resulting ensemble-guided, LLM-augmented logistic-LASSO outperformed the threshold-based method.
We evaluated this approach on both diabetes and peripheral artery disease (PAD), two phenotypes with distinct clinical presentations and documentation patterns. Ensemble disagreement proved to be a phenotype-agnostic and effective criterion for targeted augmentation. Compared with full cohort augmentation, this strategy prompted the LLM for only 10\% of patients on average, yet achieved comparable—or occasionally superior—performance, delivering significant gains in cost-efficiency, scalability, and sustainability.
Finally, we incorporated a human-in-the-loop (HIL) mechanism for targeted label correction and identification of high-quality examples for LLM self-improvement. Iterative fine-tuning with expert-reviewed cases consistently improved sensitivity, negative predictive value, and overall accuracy across development, internal validation, and external validation cohorts. Together, these findings demonstrate that targeted, uncertainty-guided LLM integration can deliver high performance while preserving portability across settings.
Key contributions include: (1) a transparent, interpretable, and uncertainty-aware method for integrating LLMs into phenotyping pipelines; (2) an ensemble disagreement metric as a scalable and objective patient selection strategy for augmentation; and (3) a HIL-driven self-improvement process to refine performance. Limitations include the cost of LLM inference, the site-specific nature of self-improvement gains, and the need for adaptation to new clinical domains. Overall, this framework offers a practical, clinician-friendly pathway for enhancing disease detection from EHR data—balancing innovation with interpretability and adaptability
Veils of the Unseen
No full textThis thesis explores the dynamic interplay between consciousness and unconsciousness and how this relationship influences visual language in my art practice. My research investigates how unconscious impulses and conscious choices intersect in the creative process, shaping artistic expression and viewer interpretation. I examine how visual art can serve as a medium through which hidden emotional and psychological states, such as memory, desire, and internal conflict, emerge and take form. Rooted in the practice of automatism and influenced by Surrealist strategies, my work engages with theories by Jacques Lacan, particularly his notion of the gaze, and Gilles Deleuze’s concepts of extraction and isolation. These theoretical frameworks have guided my response to visual repetition and unconscious copying. My art seeks to make visible what often remains hidden and to disrupt habitual ways of seeing, encouraging audiences to reconsider what they perceive as familiar. Ultimately, this thesis offers a deeper inquiry into how visual art can be a powerful tool for accessing unconscious content and opening new possibilities for self-reflection and psychological understanding
nOt JuSt AnOtHeR 2000 wOrD wRiTtEn MfA tHeSiS
No full textThis thesis challenges the standardized academic expectations placed upon studio-based MFA programs, using the process of writing itself as a site of institutional critique. Emerging from a 27,000-word working draft titled Aesthetics of Real, this final text adopts the permitted format of an “extended artist statement” to expose contradictions within institutional frameworks. It resists the assumption that professionalism and compliance equate to intellectual rigor, arguing that true artistic inquiry demands risk, failure, and confrontation with systemic norms. Through works like 3MUSTANGS, Facades of Dallas (Hall), SuperPOD0043: An American Painting, and Babel’n On, this thesis reveals how institutions often mistake performance for substance—and how, without critique, even care becomes a mechanism of containment. Refusing to kneel to formal assumptions, the thesis asserts that Real is not simply encountered, but made—manipulated, stretched, stressed, and revealed. Ending mid-sentence, it complies in form while refusing in spirit—forcing the archive itself to confront what it chooses to preserve as success, rigor, and Real
The Myth of Individualized Prosecutorial Enforcement
No full textThe Criminal Justice Reform Scholarship Workshop The Myth of Individualized Prosecutorial Enforcement was hosted by the Deason Criminal Justice Reform Center on March 8, 2023. The workshop featured Prof. Justin Murray, New York Law School
The Tenth Anniversary of Marriage Equality: How Traditional Marriage Law Led to Constitutional Protection for Same-Sex Marriage
No full textThis essay explores how the history of interstate marriage recognition law was central to the Supreme Court’s recognition of constitutional protection for the right of same-sex couples to marry. Prior to the 1990s, there were essentially no laws on same-sex marriage in the United States. That changed in the 1990s, when the Hawaii Supreme Court issued a ruling in Baehr v. Lewin that made authorization of same-sex marriage seem inevitable in that state. The threat fueled the adoption first of the federal Defense of Marriage Act (DOMA) in 1996 and then mini-DOMAs in most states. Those laws were broad and unprecedented. DOMA singled out one type of marriage for nonrecognition under any federal law or program, while most federal laws do not provide definitions of marriage, even where marital status is highly relevant such as in eligibility for spousal Social Security benefits. Rather, Congress has traditionally deferred to state law determinations of personal status when applying federal laws. State DOMAs prohibited the celebration of same-sex marriages within their borders and recognition of those celebrated elsewhere. These laws were also unprecedented in that they barred interstate recognition of all same-sex marriages without any consideration of the common-law rules that usually drove such determinations. The departures from the usual approach to federal-state and interstate marriage recognition provided a doctrinal hook for constitutional protection. Although the Supreme Court recognized the right of same-sex couples to marry in Obergefell v. Hodges in 2015, the key inroad was its decision in Windsor v. United States, two years earlier. In Windsor, the Court struck down the federal law provision of DOMA based on the idea that discrimination of an unusual character raises the specter of animus—and animus cannot be the sole justification for a valid governmental action, even when the targeted group has not been recognized as a suspect or quasi-suspect class. The same argument was then used to attack the state DOMAs. The “discrimination of an unusual character” was the legislatures’ blunt and categorical non-recognition of same-sex marriages, despite strong histories in most states of granting recognition to marriages that were validly celebrated elsewhere despite strong opposition within the state. The opponents of same-sex marriage unwittingly undermined their own cause by enacting such unprecedented and unforgiving nonrecognition laws
Global Pathways: Exploring Careers in International Law
No full textThe State Bar of Texas International Law Section and The SMU Rowling Center of Business Law & Leadership Present Global Pathways: Exploring Careers in International Law. Discover your future in international law and connect with experts who can guide you on the path to success! Panelists: Cynthia Rigney, BT Americas, Inc. Daniel Avila, Reed Smith LLP., van Castaneda, Cacheaux, Cavazos & Newton Eric Hinton, The SMU Rowling Center for Business Law & Leadership Demetra Koelling, On2It Cybersecurity (Moderator
Mobile Computer Vision Application for Agricultural Disease Detection of Pepper Diseases using Two-Stage Deep Learning System
No full textPlant diseases pose a significant threat to food security, particularly in developing countries where farmers often lack the resources and infrastructure for early detection. In nations like Mexico and the Dominican Republic, the spread of harmful plant diseases impacts key agricultural commodities, such as habanero peppers, leading to substantial yield losses. This study presents a computer vision system based on Convolutional Neural Networks (CNNs) and an object detection model (YOLO) to help farmers detect pepper diseases efficiently. The system uses a two-stage approach: YOLOv11n first detects pepper leaves in images, then a lightweight MobileNetV3Small model classifies whether the detected leaves show signs of disease. The MobileNet model has been fine-tuned specifically for pepper disease classification and optimized for deployment on smartphones, offering a cost-effective and accessible solution. The mobile application operates in real-time and offline, maintaining functionality even in areas without network connectivity. Farmers can diagnose diseases by capturing images of affected leaves, enabling early intervention and improved crop management. By leveraging deep learning (DL) for on-device disease detection, this approach can enhance agricultural productivity, reduce economic losses, and contribute to food security. This work demonstrates the potential of mobile DL solutions to transform small-scale farming through accessible, and scalable disease diagnosis technology
GPU-Based Visual Effects System
No full textThe objective of my thesis is to create a robust and efficient VFX system that can be used to edit and add particle effects to games. This system utilizes a compute shading pipeline to simulate millions of particles in real time. The behavior of particles is widely customizable through many different properties which can be manipulated changed over the lifetime of particles and introduce procedural randomness. There are many ways to customize the motion of the particles with various forces and collision. Additionally, particles can be rendered as billboarded quads, full meshes or partial meshes with different settings to further customize appearance. The behavior of particles is entirely driven by the particle emitter they are spawned from. A collection of one or more of these particle emitters make up particle effects, which is what the end user will control. All particles are simulated entirely on the GPU allowing for all particles to be updated in parallel with minimal data copied from the CPU to GPU
Voxel-Based Physically Simulated Particle System for Realistic Smoke Effects with Responsive Interaction to Dynamic Objects
No full textThe Voxel-Based Physically Simulated Particle System is designed to provide a high-performance, interactive, and visually realistic smoke simulation for real-time applications. This system replicates the dynamic smoke effects seen in Counter-Strike 2, ensuring high visual fidelity and interactivity through advanced rendering and simulation techniques.
The system leverages compute shader-based ray marching for volumetric rendering, enabling physically-based lighting simulation with realistic shading and soft shadowing effects. To achieve efficient and scalable smoke propagation, it employs Intel SIMD-optimized flood fill algorithm, significantly accelerating the expansion calculations. With over 10,000+ dynamically interacting particles, the system maintains a stable 60FPS across a wide range of hardware, ranging from systems equipped with integrated graphics and entry-level CPUs (e.g., Intel UHD Graphics, Ryzen 3) to machines with dedicated GPUs and high-performance processors (e.g., RTX 3080+, Intel i9, or AMD Ryzen 9).This Technical Design Document (TDD) serves to define the system\u27s technical framework, outlines its core algorithms and optimizations, and provides guidelines for implementing the whole system