22689 research outputs found
Sort by
Quantization and Pruning of Convolutional Neural Networks for Efficient FPGA Implementation of Digital Modulation Detection Firmware
Get PDFAutomatic modulation detection is an important function of communications systems. Commonly found in software defined radios, it enables radio receivers to interpret multiple and potentially changing modulation types without needing manual input from the user. Due to vastly increasing performance, many modern systems are moving away from the traditional two-stage process of feature extraction and classification; instead, a neural-network based system (also known as deep learning) is being utilized with increased speed and virtually no loss in accuracy. These implementations, when placed on hardware or on a Field Programmable Gate Array (FPGA), provide the fastest performance; but until recently the barrier of entry has been the size of the neural networks and the infeasible amount of resources they would need to occupy on the FPGA fabric. This thesis explores the effects of both quantization and pruning on convolutional neural network models of various sizes while maintaining high classification accuracy for the digitally modulated signals generated. In this thesis, a framework is proposed for the generation of the signals, models, and hardware estimation to serve as a guide for efficient deep learning implementations of models intended to fit on hardware with limited resources. The results demonstrate tradeoffs and design considerations that balance performance and implementation size for engineers aiming to implement a deep learning-based automatic modulation detection scheme on FPGAs
ASSESSING THE EFFECT OF COMPUTABLE PHENOTYPES IN PREDICTING HEALTHCARE UTILIZATION AND POTENTIAL RACIAL DISPARITIES IN IDENTIFYING TYPE 2 DIABETES POPULATION
Get PDFStatement of the Problem
Computable phenotype definitions are often used to identify population with the chronic conditions, such as Type-2 Diabetes (T2D) using electronic health records (EHRs) and administrative claims (claims) data. The variations in phenotypic characteristics and their performance in identifying patients with T2D may have downstream impact for clinical research, epidemiological studies, and population health management efforts. To bridge these gaps, our study aims to understand the complexities of T2D computable phenotypes and their impact on predicting healthcare utilization and racial disparities.
Research Design and Methods
We conducted a retrospective data analysis using Johns Hopkins Medical Institute (JHMI) EHR and Johns Hopkins Healthcare (JHHC) claims data of adult patients during 2017 and 2019. The study included five computable phenotypes – 4 published, commonly available definitions, and 1 unpublished definition from experts at Johns Hopkins. Our study methods were multi-fold such that we 1). analyzed the effect of induced data quality issues using EHR and claims data, 2). Conducted logistic regression models during the 3-year study duration to assess performance of phenotypes in predicting healthcare utilization, such as inpatient (IP) and emergency room (ER) visits, and 3). analyzed the population distribution and predictive modeling techniques stratified by racial groups across phenotypes.
Summary of Results
The variations in phenotypic characteristics resulted in identifying different populations with T2D across phenotypes during our study period. Our methods resulted in statistical differences in population demographics, comorbidity scores, utilization of IP and ER visits, performance of predictive models across phenotypes and by racial groups. Additionally, the assessment of data quality issues of completeness, accuracy and timeliness showed significant impact in capturing (or lack of) populations identified.
Conclusion
It is important to understand factors that contribute to the complexity and performance of computable T2D phenotypes when identifying populations of interest using retrospective EHR and claims data. Our results showcase a simple and pragmatic yet challenging approach that population health researchers and key stakeholders must consider before implementing computable phenotype(s) to identify, enroll, and analyze T2D populations of interest
CHARACTERIZATION OF CELLULAR SENESCENCE ASSOCIATION WITH PANCREATIC FUNCTION USING MULTIPLEXED MOLECULAR LABELING IN HUMAN PANCREATIC TISSUE
No full textIn recent years, there has been a significant increase in age-related diseases due to the
improvement in average lifespan. Aging is associated with a myriad of morphological and
pathological changes in various organs, including the pancreas. These changes predispose
individuals to a spectrum of aging-related diseases such as diabetes, pancreatic ductal
adenocarcinoma, and pancreatitis. Cellular senescence, defined as a state of permanent cell
cycle arrest induced by various stressors, plays a critical role in both physiological aging and
pathological conditions.
Advanced multiplex immunofluorescence techniques have been leveraged to detect multiple
senescence-associated and functional markers concurrently in pancreatic tissue sections. Our
findings highlight the spatial distribution of senescence markers like p16 and 53BP1 and their
association with insulin secretion in the islets. Through optimized staining protocols and
imaging analyses, our work provides insights into the senescence landscape of the pancreas,
highlighting the complex interplay between cellular aging and pancreatic function
Exploring disparities in quality-of-life among working women with menorrhagia/dysmenorrhea
No full textIntroduction: Menorrhagia (heavy or prolonged menstrual bleeding) and dysmenorrhea (painful menses) are two of the most common gynecologic disorders affecting people who menstruate. Menorrhagia/dysmenorrhea have known quality-of-life (QOL) and work-related impacts, which may be potentially exacerbated for women with lower socioeconomic status and women of color, who are disproportionately represented in low-wage jobs. However, there has been little exploration on potential disparities in QOL based on socioeconomic status, occupation, or race and ethnicity. Guided by the Wilson and Cleary health-related QOL model and the social-ecological model of health, we conducted an exploratory sequential mixed methods study to explore disparities in QOL among working women with menorrhagia/dysmenorrhea, with a focus on identifying relevant occupational variables and structural and social factors that contribute to inequities.
Methods: A literature review on the work-related impacts of menorrhagia/dysmenorrhea was conducted to identify gaps in the literature and guide our study development. In the first stage of the study, we conducted 12 qualitative one-on-one interviews using a semi-structured interview guide to explore women’s experiences with menorrhagia/dysmenorrhea. We conducted a thematic analysis of the interview data to identify the most pertinent themes and refine the development of the survey instrument used in the second stage of the study, in which we administered a quantitative, cross-sectional survey to 160 women with menorrhagia/dysmenorrhea. We conducted a statistical analysis to identify demographic, socioeconomic, and occupational variables with significant associations on QOL, symptom severity, and work-related impacts. The qualitative and quantitative results were analyzed collectively to understand the structural and social factors associated with QOL disparities from both the lived experienced and supporting statistical data.
Results: Five themes were identified from the thematic analysis of the interview data: 1) Healthcare journeys, 2) Navigating the workplace, 3) Gendered challenges, 4) Stigmatization, and 5) Recommendations for healthcare and workplace support. Participants described poor healthcare experiences, stigmatization from healthcare providers and employers, and challenges in the workplace due to insufficient systems to support women with menorrhagia/dysmenorrhea. The survey results identified several associations between poor healthcare experiences and inequitable working conditions that contribute to disparities in QOL, symptom severity, and work-related impacts, and women with lower socioeconomic status and women of color reported worse health-related QOL and menstrual symptom severity. Our integrated analysis highlights the need to implement organizational and policy-level interventions to alleviate socioeconomic, racial, and occupational inequities.
Conclusion: We identified several disparities in QOL among working women with menorrhagia/dysmenorrhea due to inequitable healthcare access, poor working conditions for women with chronic conditions, and stigmatization in the workplace and health system. We recommend future exploration of policies and interventions to improve screening and treatment of menstrual conditions, enact paid leave policies and workplace accommodations for workers with menorrhagia/dysmenorrhea, and increase Diversity, Equity, and Inclusion efforts to increase the representation of women in both the medical field and in leadership positions
REVIVING THE ECOPATENT COMMONS: GIVING CLIMATE PATENTS A SECOND LIFE
No full textThis paper proposes the Climate Tech Patent Commons (CTPC), a strategic policy proposal to expedite the deployment of crucial technologies to combat climate change. Over the years, patents have increasingly become critical business assets, not just for their protective value but also as tools for securing investments and enhancing corporate valuations. This growth has led to complex landscapes of overlapping patents, known as patent thickets, and the emergence of patent trolls—entities that exploit the patent system for profit without contributing to actual innovation.
The CTPC seeks to address these challenges by creating a managed platform for patent sharing. This initiative organizes and categorizes patents to enhance access and utility, simultaneously reducing the barriers erected by patent thickets and mitigating the obstructive behaviors of patent trolls. By lowering patent maintenance fees and offering incentives such as expedited patent processing, the CTPC encourages robust participation from both the public and private sectors. This cooperative approach facilitates quicker commercialization of climate technologies and fosters a collaborative environment that promotes ongoing technological innovation. Ultimately, the Climate Tech Patent Commons is envisioned as a transformative tool, leveraging intellectual property to drive environmental sustainability and innovation more effectively
A Computational Study of Corrosion in Nickel-Titanium Alloys in Saltwater Conditions
Get PDFCorrosion is a naturally occurring process that can be beneficial when controlled, yet can cause immense damage when allowed to run unchecked. Technologies have been developed to combat unwanted corrosion, with one of the more prominent solutions being thin film coatings of corrosion-resistant materials such as nickel titanium (NiTi). In addition to the corrosion-resistant nature of this alloy, NiTi possesses the unique properties of superelasticity, biocompatibility, and a reversible phase transformation that allows for self-healing. This combination of factors has led to extensive experimental studies on the corrosion of NiTi; however, the cost of materials and analytic equipment can be a sufficient burden to make a computational analysis via density functional theory (DFT) as a viable alternatve. DFT has been used to characterize properties of NiTi in the past, but few DFT studies of corrosion have been performed with this approach.
This thesis attempts to fill this gap in knowledge by simulating the adsorption of two ionic species (Cl and OH) found in salt water onto the low-index surfaces of NiTi and calculating how the adsorption of these species affect the energy landscape of the surface. We begin with pristine surface terminations on the (110), (100), and (111) planes of NiTi and demonstrate that our methodology can replicate existing literature values of the electron density of states and surface energies on this alloy. Cl and OH adsorbates are then introduced into the system and allowed to relax onto the NiTi surface. We show that OH will tend to bind to all surfaces with greater strength than Cl and also show that both adsorbates will prefer to bind at sites that maximize bonding to Ti atoms over Ni atoms in the surface. The addition of either adsorbate reduces the surface energy of NiTi, which is supported by Gibbs adsorption theorem. Overall, we demonstrate a methodology to study corrosion that can be adapted to study other surfaces of NiTi, other materials, or other adsorbates to recreate different environments and solutions
Longitudinal Patterns and Predictors of Hazardous Alcohol, Opioid, and Stimulant Incidence and Cessation among Female Sex Workers Living with HIV in South Africa
Get PDFBackground: Female sex workers (FSW) in South Africa are disproportionately impacted by HIV and substance use disorders. Substance use has been associated with poor HIV treatment outcomes, necessitating exploration of polysubstance use patterns and predictors among FSW living with HIV.
Methods: Data on substance use and relevant covariates were obtained for 777 FSW randomized to the Siyaphambili HIV treatment strategies trial implemented through TB HIV Care. FSW were screened for recent marijuana, opioid, stimulant, and hazardous alcohol use at enrollment and semi-annually for 18 months from June 2018 to January 2022. Individuals were assessed for substance use initiation/cessation after enrollment, and Kaplan-Meier plots and lasagna plots visualized these trends. Cox proportional hazards models assessed baseline predictors of substance use incidence and cessation.
Results: Opioid use and abstinence appeared more consistent over the study period than hazardous alcohol and stimulant use. Exhibiting symptomology of hazardous alcohol use at first visit was inversely associated with incident opioid use (aHR = 0.13, 95% CI = 0.04, 0.43), while reporting opioid use at first visit was inversely associated with incident hazardous alcohol use (aHR = 0.32, 95% CI = 0.16, 0.61). Reporting opioid use at first visit also predicted incident stimulant use (aHR = 4.04, 95% CI = 2.38, 6.86), and reporting stimulant use at first visit predicted incident opioid use (aHR = 2.62, 95% CI = 1.14, 6.01).
Discussion: Opioid use was a predictor of incident stimulant use and vice versa, suggesting that concurrent opioid and stimulant use is common among FSW. These findings highlight a need for multipronged substance use interventions for FSW that integrate harm reduction approaches to opioid and stimulant use. Further investigation into these complex polysubstance use patterns and HIV treatment outcomes could inform substance use interventions and HIV care delivery strategies tailored to FSWs’ preferences and needs
ADVANCEMENTS FOR DIFFERENTIAL METHYLATION ANALYSIS: SINGLE-CASE OPTIMIZATION, METHOD EVALUATION, AND NANOPORE SEQUENCING APPLICATIONS
Get PDFDNA methylation, a major epigenetic modification, plays a crucial role in regulating gene expression, embryonic development, and genome stability. Dysregulation of DNA methylation has been associated with numerous diseases, emphasizing the importance of understanding its functional role. This thesis aims to advance DNA methylation analysis by developing and evaluating computational methods for identifying differentially methylated regions (DMRs) using both traditional and emerging sequencing technologies.
The first part of this thesis focused on optimizing DMR detection for single-case analyses against multiple controls, addressing challenges in unbalanced study designs common in clinical and rare disease contexts. Addressing issues with smoothing performance and coverage outliers, we introduced solutions such as filtering low-coverage CpGs and employing robust variance estimation. Despite fewer putative and significant DMRs detected, our methods yielded robust and accurate results.
The second part of this thesis provided a comprehensive evaluation of two widely used DMR detection methods, BSmooth and Dmrseq. While both methods showed proficiency in diverse scenarios, they exhibited unique strengths and limitations. BSmooth effectively controlled the Family-Wise Error Rate (FWER) but might overlook smaller effect sizes. Dmrseq, on the other hand, handled sparse data well but required refinement in error control. Our proposed method, BSmooth with pooled regions, offered a balanced approach for error control, effectively capturing both substantial and subtle differential methylation regions without significantly increasing false positives.
In the last part of the thesis, we explored the potential of nanopore sequencing for DNA methylation analysis, focusing on quality control and DMR identification in Dnmt1 knockout mice. Despite its advantages, nanopore sequencing introduced challenges related to data processing and interpretation. Our analysis of the impact of thresholding on nanopore methylation analysis using \textit{Kmt2a} nanopore data processed by modbam2bed revealed that sequencing errors, typically associated with low methylation scores, exhibit a random distribution in nanopore methylation data. Differential methylation analysis of \textit{Dnmt1} knockout nanopore data identified 10,699 putative DMRs, predominantly hypomethylated, spread across chromosomes 1 to 19 with chromosome 2 having the highest number of DMRs and chromosome 19 having the fewest. Notably, significant DMRs often corresponded to pronounced methylation level changes, suggesting possible regulatory mechanisms.
Overall, this thesis contributes to DNA methylation analysis by improving and evaluating existing methodologies. By addressing challenges and refining approaches, we improve DMR detection, which has implications for understanding disease mechanisms and identifying biomarkers. Additionally, by tackling challenges with emerging sequencing technologies like nanopore sequencing, we expand the toolkit available for studying DNA methylation dynamics across various biological contexts
Constructing More Realistic Models of Biogeochemical Processes: Key Considerations for the Cryptic Sulfur Cycle, Optics and Beyond
Get PDFOcean biogeochemical models are heavily reliant on numerous uncertain parameters and processes, making it crucial to improve our understanding and reduce uncertainties in these areas to enhance the models' accuracy. My thesis focuses on two primary research questions: 1. Do we have all the necessary processes? If not, what additional ones should we/can we add? 2. Can we constrain how those processes work? Three projects were done in order to address these questions. In the first project, we tried to construct a model including the cryptic sulfur cycle in Chesapeake Bay. However, the representations of particle sinking, burial, dissolved organic matter, nitrification and light attenuation all have more significant impacts on model skill than the inclusion of sulfur cycling. In the second project, we tried to study role of colored dissolved organic matter (CDOM) in coastal ocean hypoxia. The results suggest that the impact of CDOM is region-specific and multifaceted. In the upper Bay, the removal of CDOM reduces light limitation, thus promoting increased productivity, resulting in the generation of more detritus and burial, which, in turn, contributes to elevated levels of hypoxia. As we transition to the middle and lower Bay, the removal of CDOM can cause a decline in integrated primary productivity due to nutrient uptake in the upper Bay. Finally, we have begun a further investigation on distributional characteristics and sensitivities to environmental drivers of particulate inorganic carbon (PIC) versus particulate organic carbon (POC) in both remote sensing observations and models. Current findings show that different biogenic carbon species respond differently to variations in environmental drivers. PIC appears to be less sensitive to iron compared to POC. Further investigation for whether Earth System Models are capturing these different changes for different carbon species is needed for better modeling marine biogeochemical cycles since we have also identified limitations in some widely used Earth System Models, including the Community Earth System Model from the National Center for Atmospheric Research, as they struggle to simulate PIC well
Predictive Simulations of Erosion in Metals and Graphite Induced by High-Velocity Particle Impact
Get PDFParticles in a solid rocket exhaust plume can cause significant erosion damage to impinged surfaces, such as rocket nozzles and launch pad structures. To design more robust systems, computational methodologies utilizing both finite element methods (FEM) and smoothed particle hydrodynamics (SPH) have been developed to predict the steady-rate erosion associated with the impact of alumina particles. In this study, simulations focus on two ductile materials, 4340 steel and Ti-6Al-4V, and a semi-brittle material, EDM-3 graphite. The impact of particles with velocities ranging from 400–900 m/s and impact angles of 10–90 degrees were examined. Both liquid and solid-state particles with spherical and cubical geometries were considered. A novel modeling approach was employed, utilizing periodic boundary conditions of the target to represent an infinitely large plate. Furthermore, a sufficient set of randomly impacting particles was employed to reach steady-rate erosion from multiple particle impacts on the exposed surface. The results showed that 4340 steel exhibited superior erosion resistance compared to the two other materials studied. Ti-6Al-4V had approximately twice the erosion rate of 4340 steel, while EDM-3 graphite had approximately five times the erosion rate. Also, the use of both FEM and SPH simulations provided valuable insights into each approach's advantages and limitations in predicting the erosion rates of materials. The models were validated against experimental findings in the open literature and demonstrated good agreement. Notably, the SPH approach was found to be more consistent with experimental findings in terms of absolute erosion rate. This study establishes a solid foundation for simulating mechanical erosion for various particle and target material combinations. The developed methods can be applied to predict erosion in other applications, such as jet engine particle ingestion, rain erosion of aircraft surfaces, and particle damage in oil pipelines. This study's findings can contribute to developing more robust systems capable of withstanding particle erosion in challenging environments