Washington University Medical Center
Washington University St. Louis: Open ScholarshipNot a member yet
26344 research outputs found
Sort by
Development of a Multispectral Collagen detection dermatology device for delineation of Non-Melanoma Skin Cancers
No full textTo diagnose and manage skin disease, clinicians primarily use visual inspection aided with dermatoscopes, which are cross-polarized magnifiers. In recent years, digital cross polarized multispectral imaging systems have been developed to quantitatively assess skin disease. For example, systems like the one developed by Yaroslavsky et al., which utilize red and blue filters, can non-invasively assess the collagen network in the dermis(Yaroslavsky et al., 2003). Visualization of collagen network disruption can make it easier to visualize the boundaries of skin cancers like keratinocyte carcinoma, and therefore easier to excise completely. Current systems use external bandpass filters in front of cameras to achieve multispectral cross-polarized imaging, and this adds complexity, bulkiness, and extended exposure time during which motion artefacts are more likely.
The proposed project consists of developing a handheld, cross-polarized digital system that displays the collagen network of skin. The ability of the device to perform collagen network detection will be validated by imaging healthy subjects with varying skin pigmentation to assess healthy skin collagen network, and imaging skin cancer patients undergoing Mohs surgery
Implementation of Machine Learning and Large Language Models for High Entropy Alloys
No full textHigh-entropy alloys (HEAs) are a relatively recent class of materials formed by alloying five or more elements in near-equimolar ratios. Their promising mechanical and thermal properties make them promising for a wide range of applications, however, the vast compositional space presents a challenge for efficient exploration. As a result, robust and structured databases are critical for guiding HEA research and accelerating materials discovery. In this project, a computational pipeline was developed to leverage large language models (LLMs) for extraction of compositional data and relevant properties from existing HEA literature. The program integrates structured prompt design, parallel execution, and deterministic post-processing to compile mined data into one large database. Performance evaluation demonstrated high consistency across repeated queries, and the full pipeline achieved a total runtime of under 1.5 minutes to process fifty papers. This work demonstrates the feasibility of using LLM- assisted workflows to rapidly construct materials databases from unstructured literature, offering a scalable approach for future materials informatics and data-driven alloy design efforts. Future work will focus on incorporating additional validation mechanisms and creating an interactive website to view the compositional data
MEMS 4110: ASME Waste Collection Challenge
No full textThe Efficient Jolley Trash Grabber (EJTG) is a teleoperated vehicle adherent to the modified rules of the 2025-26 ASME Student Design Competition, hosted in MEMS 4110: Mechanical Engineering Design Project. The vehicle design was informed first by understanding the problem of waste collection. Patents, existing devices, user needs, and relevant codes and standards were reviewed to survey potential design approaches. The concept for the vehicle was generated by setting functionality, and brainstorming solutions to accomplish each goal. A mock-up vehicle was created using available materials, and four complete concepts were drawn each with a different design. Rigorous methods were used to determine for priorities the ideal design, and a concept was chosen which included flexible grippers and a trap-door to store garbage. Three engineering models analyzing required torque to rotate a weighted arm, battery capacity, and gripper distance informed the component selection and dimensional design of the initial prototype
MEMS 4110: Parachute Snatch Force Tester
No full textThis report discusses the design, development, and testing of a parachute snatch force tester for MEMS 411 Senior Design. The device is for the WURocketry Team, and aims to measure the peak force exerted on the rocket bulkhead during parachute ejection. This peak force is known as the snatch force, and is incredibly important when it comes to maintaining safety, improving performance, and material selection during the rocket’s production process. Current methods for testing and recording this snatch force are limited, so our team aimed to create a simple and cost effective solution. Our design utilizes a vertical support that stores elastic energy in a bungee cord. When released, the stored energy launches the bulkhead downward until the attached shock cord fully tightens. The shock cord is attached to a load cell which is connected to an HX711 amplifier. The load cell then measures the peak force at the moment the shock cord experiences full tension. Three performance goals were set at the beginning of this project. The first was to measure bulkhead acceleration values within 5% of flight data. This goal could not be fully met due to a malfunction with our accelerometer. The second goal was to measure force values within 10% of calculated values from models. While we were able to obtain consistent force values from trial to trial, they were not within 10% of the calculated values. The final goal was to have the device operate 10 times in a row with less than 5 minutes between each trial. This goal was successfully met, and there were no concerns when running the device through multiple trials
The Development of Deep-Learning-Based Automatic Multi-Organ Segmentation Models from CT Images and their Clinical Evaluation
No full textSegmentation of organs and lesions in medical images is a critical component of clinical workflows, supporting tasks such as diagnosis, prognosis, and treatment planning. However, manual segmentation is labor-intensive, time-consuming, and prone to human error, motivating the development of automated segmentation methods. Among these, deep learning–based approaches have gained substantial attention in recent years due to their ability to learn rich, data-driven representations. Two widely used architectures are convolutional neural networks (CNNs) and vision transformers (ViTs). Despite their success, these models face challenges in multi-organ segmentation from volumetric medical images due to inherent architectural limitations. Furthermore, their clinical deployment is hindered by performance degradation when applied to unseen domains or imaging modalities—a problem known as domain shift. Domain shift substantially limits the direct transferability of segmentation models across datasets without explicit adaptation. Recently, Medical Vision Foundation Models (Med-VFMs) have emerged, offering strong performance by leveraging prior knowledge acquired through large-scale self-supervised pretraining. These models can be adapted to segmentation tasks via fine-tuning on a small set of labeled samples from the target domain. However, existing studies have not investigated effective strategies for selecting these samples, leaving open the question of how to adapt Med-VFMs most efficiently to new domains. To address these challenges, this dissertation introduces four methods designed to achieve two specific aims: (i) developing automatic multi-organ segmentation models from CT images and (ii) cross-domain adaptation and evaluation. (1) First, attention mechanisms and multi-scale convolutional kernels are integrated into CNNs to improve their ability to capture both global contextual information and multi-scale features, thereby enhancing performance in segmenting organs with substantial inter-patient variability in shape and size. (2) Second, large-kernel convolutional layers are incorporated into ViTs to enrich their capacity for capturing fine-grained localization cues, which in turn improves the delineation of adjacent organs and the segmentation of small structures by more accurately identifying boundary regions. The inherent architecture of ViTs facilitates the utilization of global information, thus further enhancing their ability to preserve anatomical coherence when segmenting organs with considerable shape variability. (3) Third, an Active Source-free Cross-domain and Cross-modality Adaptation framework is proposed to adapt segmentation models across different domains and modalities. This method employs active learning (AL) to selectively query informative target-domain samples using a novel Active Test-Time Sample Query strategy to guide model optimization. (4) Finally, an Active Selective Semi-supervised Fine-tuning approach is proposed to efficiently adapt Med-VFMs for volumetric medical image segmentation. This method also leverages AL to identify the most informative target-domain samples for fine-tuning without requiring access to the original source data, thereby maximizing performance with minimal annotation cost
Investigating the role of Cornichon-homolog 3 (CNIH3) in opioid use, contributing risk factors, and associated neural substrates
No full textOpioid misuse remains rampant as potent synthetic opioids such as fentanyl flood the market. Large-scale genetic tools like the GWAS identify previously unrecognized targets and biomarkers in opioid misuse with the hopes of combating the opioid epidemic and opioid use disorder (OUD). One such target is the AMPA receptor (AMPAR) auxiliary protein Cornichon Homolog-3 (CNIH3), which determines AMPAR subunit composition and kinetics. Though CNIH3 was identified as a gene of interest in OUD, its role in opioid use and accompanying risk factors has not been studied. Using mice with a CNIH3 deletion, we assess the role of CNIH3 in risk factors for opioid use, cognition, and opioid use itself. We find that CNIH3 deletion moderately impairs spatial memory, reward-cue association, and reversal learning. CNIH3 deletion also impairs fentanyl-cue association and blunts fentanyl intake during IVSA. We use principal component analysis to pinpoint the dimensions in which CNIH3 deletion impacts behavior in an unbiased manner. Additionally, we identify in previously published human data that single-nucleotide polymorphisms are more protective against progression to daily opioid use in women than in men, suggesting a potential sex-specific role of CNIH3. Next, we use Western Blotting to determine AMPAR subunit activation and expression in key opioid-related brain regions: the medial prefrontal cortex (mPFC), hippocampus (HPC), and nucleus accumbens (NAc), in naïve and IVSA-trained male and female CNIH3 KO and WT mice to assess CNIH3-driven regulation of glutamatergic transmission in the context of opioid learning. Preliminary data suggest no basal differences, but sex-specific activation of GluA1 in the mPFC after IVSA. These findings highlight an important role of CNIH3 in opioid use through learning and memory processes and AMPAR-mediated plasticity that differ between males and females
Time-Resolved Ultrafast Infrared Spectroscopic Interrogation of Intramolecular Reaction Coordinates
No full textLight-induced chemical transformations are practical across many chemical and biological systems. From intramolecular hydrogen-bonding coordinates to reactions which alter the carbon framework of a molecule, the application of light can induce conformational changes that can subsequently be harnessed for a variety of tasks. However, molecular-level mechanistic insight about photo-induced chemical reactions remains difficult to acquire using current methodologies. In this dissertation, several intramolecular reactions are investigated and characterized, ranging from ground- and excited-state intramolecular proton transfers to the photochemical synthesis of a complex organic cyclization structure. The reaction coordinates are directly interrogated by utilizing ultrafast infrared spectroscopic techniques, such as time-resolved transient absorption and two-dimensional spectroscopy, which monitor vibrational motions within molecules that are highly sensitive to structural changes over the course of a reaction. First, a study on the intramolecular H-bond dynamics in a set of β-diketones based on the framework of acetylacetone was performed. For the compounds with shorter H atom donor-acceptor distances, which correspond to stronger H-bonding interactions, the calculated H atom potentials are quite soft and result in a red-shifted OH stretch vibrational mode frequency and H atom dislocation after vibrational excitation of this mode. Coupling between the OH stretch to many low-frequency normal mode coordinates was shown to contribute to the broad experimental OH stretch vibrational signature. Polarization-dependent transient absorption studies suggest significant isotopic-dependent differences on the orientational relaxation timescale, occurring around ~600 fs and ~2 ps for the OH and OD stretching regions, respectively. These timescales were interpreted as pertaining to H/D atom transfer events influenced by the presence of intramolecular structural rearrangements. Next, the proton transfer dynamics upon electronic excitation of 3-hydroxyflavone (3HF) and 3-hydroxy-2-(thiophen-2-yl)chromen-4-one (3HTC) were studied using transient infrared (TRIR) spectroscopy in a relatively nonpolar solvent (CDCl3). Rapid proton transfer dynamics (\u3c100 fs) were determined to occur in both systems, shortly followed by vibrational relaxation of the tautomeric conformation of the electronic excited state. The faster vibrational relaxation timescale of 3HF (~1 ps) compared to that of 3HTC (~3 ps) was attributed to the greater number of vibrational modes, with coupling to low-frequency vibrational modes invoking motion along the H-atom transfer coordinate observed for the OH bend in both the ground and electronic excited state. Finally, transient infrared spectroscopy was applied to a more complex chemical reaction, tracking the kinetics and dynamics of the cyclization of an intramolecular [2+2] cycloaddition prompted by the application of visible light. Within ~300 ps, vibrational signatures corresponding to formation of the cyclized product were observed. Application of a four-step sequential reaction model via global lifetime analysis returned evolution-associated spectral profiles, the first two of which are hypothesized to derive from fast vibrational relaxation from the initial excited electronic state (~160 fs) and internal conversion (~3 ps). Following these steps, a pivotal intermediate state or species appears to form on a ~30 ps timescale, possibly caused by an intersystem crossing into the triplet manifold. This intermediate then undergoes a rate-limiting transformation, ultimately resulting in product formation on a timescale of ~260 ps. With this information regarding the occurrence of rapid product formation and possible mechanistic insights, practical measures can be taken in macroscopic synthetic procedures to optimize the efficiency of product formation for this and similar chemical reactions
Development of Turbulence Models for Prediction of Hypersonic Flows Using Compressibility Corrections and Uncertainty Quantification With Emphasis on the Wray-Agarwal Model
No full textAccurate prediction of high-speed compressible turbulent flows remains one of the most significant challenges in Computational Fluid Dynamics (CFD), especially in the realm of hypersonic vehicle design. These flow regimes involve significant aerodynamic heating and shock-wave boundary layer interactions (SBLI) which remain a significant challenge for the current CFD prediction capabilities. While Direct Numerical Simulation (DNS) and Large-Eddy Simulation (LES) provide high-fidelity solutions, their computational demands limit their usability, making the Reynolds-Averaged Navier-Stokes (RANS) equations an efficient alternative for many industrial aerodynamic applications. Traditional RANS techniques, while computationally efficient, often provide limited accuracy in high-speed flows due to the presence of strong compressibility effects and density variations. This dissertation contributes to the advancement of one-equation Spalart-Allmaras (SA) and Wray-Agarwal (WA) linear eddy viscosity turbulence models, with emphasis on the high-speed flow application of the WA model. All turbulence models and their proposed compressibility corrections are implemented in the open-source CFD solver Stanford University Unstructured (SU2). Following implementation, verification and validation was conducted for many canonical benchmark cases from the NASA Turbulence Modeling Resource (TMR) website to assess the model’s accuracy and reliability. The proposed WA model modifications include incorporation of the Catris and Aupoix compressible boundary layer corrections and a newly developed variable-property compressible law of the wall correction (WA-CCLoW). These enhancements are derived, and the improved WA model’s performance is evaluated on an extensive suite of experimental supersonic/hypersonic NASA benchmark cases. Furthermore, Uncertainty Quantification (UQ) techniques are employed to assess the sensitivity of the closure coefficients of the turbulence model on the solution and thereby fine-tuning the model coefficients for high-speed flow predictions. This research lays a foundational framework for the continued development and advancement of one-equation RANS turbulence models as practical and computationally affordable tools for hypersonic flow simulations. By combining the classical fluid dynamics theory with complex turbulence model development, uncertainty quantification, and sensitivity analysis, this dissertation contributes to the broader scope of advancing the RANS predictions in the field of high-speed aerodynamics
Quantum Multi-Photon Systems: Photonic Bound States and Their Applications
No full textPhotonic bound states represent a new class of quantum multi-photon states. Due to previous interactions in the quantm nonlinear medium, photons in a bound state exhibit effective attraction and propagate together as a composite entity. Their strong inter-photon correlations offer promising advantages in quantum imaging, quantum communication, and quantum computing. While two- and three-photon bound states have been observed in cold-atom experiments, efficient generation in practical systems remains a significant challenge. This dissertation tackles the challenges on two fronts. First, I investigate the generation schemes and statistical properties of two types of photonic bound states: multi-photon states arising from superradiant emission and photonic-dimer coherent states—ensembles of two-photon bound states analogous to coherent states of single photons. The correlation functions and optical coherence properties of these states are analyzed in detail. Second, I show that their intrinsic photon-photon correlations enhance performance in turbulence-free interferometry and increase two-photon excitation efficiency, enabling deeper tissue imaging in two-photon fluorescence microscopy. Overall, this work advances our understanding of multi-photon quantum systems and expands the potential applications of photonic bound states in quantum technologies