22689 research outputs found
Sort by
BREASTFEEDING SUPPORT AND MATERNAL INFANT FEEDING DECISION-MAKING AND PRACTICES IN YUCATAN, MEXICO
No full textBackground
There are many health and development benefits of breastfeeding for infants, as well as health benefits for the mother. In Mexico, the prevalence of optimal breastfeeding practices are below global targets and disparities exist.
Objectives
This research aims to understand how facility-, maternal-, and community-level characteristics influence breastfeeding practices in the first two months postpartum among women in Yucatan, Mexico.
Methods
This study used multiple methods and respondent types. Semi-structured interviews with healthcare providers (n=32), observations of breastfeeding support services (BSS) after delivery (n=10), and a content analysis of hospital infant feeding protocols and policies were conducted. Data analysis was guided by the Ten Steps to Successful Breastfeeding. Longitudinal surveys of women were conducted in hospitals within 3-days of delivery and at 2-months postpartum. Maternal-reported receipt of BSS was collected and counted via a summative score with 6 possible points. Regression analyses evaluated the associations of the BSS score and maternal characteristics, and breastfeeding practices. Semi-structured interviews were conducted with 16 women who completed the 2-month survey. Thematic analysis was completed to identify facilitators and barriers to exclusive breastfeeding at 2-months.
Results
Out of 345 women, 99.1% reported breastfeeding, 68.4% initiated within one hour of birth and 68.4% exclusively breastfed during their hospital stay. Receipt of BSS among women were low, with mean scores of 3.25 points (SD 0.99) out of 6 points. Mean BSS score for those with cesarean deliveries was even lower (2.95 points [SD 0.95]). While hospitals use rooming-in in facilities, facilitation of skin-to-skin contact was uncommon, women and infants were commonly separated after delivery, and formula was sometimes introduced for non-medical reasons. The BSS score was associated with greater odds of early initiation of breastfeeding (adjusted odds ratio (aOR)=1.73; 95% CI: 1.31, 2.29), but not with exclusive breastfeeding at 2 months (aOR=1.08; 95% CI: 0.82, 1.42). Other barriers to exclusive breastfeeding identified in qualitative interviews were perceived milk insufficiency, infant discontented behaviors, and employment/school.
Conclusions
This study informs efforts to improve provider training, breastfeeding counseling, and behavior change programming to increase optimal breastfeeding behaviors. Further research is needed to better understand inequities in receipt of hospital-based BSS, and barriers after discharge, such as access to milk pumps
INVESTIGATING THE PLASMODIUM DOSE-DEPENDENT INHIBITORY ACTION OF LUMEFANTRINE AND TAFENOQUINE
No full textMalaria remains a major global health challenge, with 597,000 annual deaths across 83 countries according to the WHO. The parasite’s ability to detoxify free heme from hemoglobin catabolism into inert hemozoin is essential for its survival, making hemozoin crystallization a critical drug target. This thesis investigates the lumefantrine and tafenoquine inhibitory mechanisms on hemozoin formation, using both biochemical assays and pulse drug assays in NF54 Plasmodium falciparum cell culture to gain insight into the stage-specific windows of action. Hemozoin crystal formation occurs in two steps - crystal nucleation which occurs in late ring stages and crystal extension that occurs in trophozoite stages. Lumefantrine potently inhibits hemozoin crystal extension, IC50 of 8.2μM, but has no measurable effect on nucleation, even at drug concentrations as high as 200μM. Despite this, lumefantrine showed robust stage-independent parasite inhibition, with IC50 values ranging from 1–3.5μM across all intraerythrocytic stages, including early rings where hemoglobin digestion and hemozoin formation are minimal suggesting non-canonical effects beyond hemozoin interference. Tafenoquine inhibited hemozoin nucleation at high in vitro 44.6μM concentrations with no appreciable impact on crystal extension. Tafenoquine also inhibited P. falciparum growth across all blood stages in pulse assays, but its potency was stage-dependent, with IC50 values increasing from 1–5μM in ring stages to 10–20μM in late trophozoites. Typically, a liver stage drug, tafenoquine, requires cytochrome P450 activation in hepatocytes, so its blood stage action may be attributed to a Plasmodium non-mitochondrial target. By exploring hemozoin-targeting mechanisms and the stage-specific effects of tafenoquine and lumefantrine, we can further understand the window of action of these antimalarials and explore potential non-canonical pathways of disruption
Chikungunya virus infection in neural cells: investiagting cell-type differences in replication dynamics and host responses
No full textChikungunya virus (CHIKV) is one of the few alphaviruses that can, on rare
occasions, cross the blood-brain barrier and directly infect neurons, though the level of virulence varies from cell to cell. In past studies, the virus has shown a clear preference for infecting astrocytes rather than neurons, which are targeted by the neurovirulent viruses most closely related to CHIKV. CHIKV virulence is dependent on the activity of the macrodomain (MD) of nonstructural protein 3 (nsP3), which can bind and remove ADP-ribose (ADPr) units from ADP-ribosylated substrates to disrupt interferon (IFN) signaling sometimes associated with ADP-ribosylation. While recent studies have investigated the
effect of variable nsP3 MD activity on CHIKV infection, research into how that variable activity impacts the preference of CHIKV to infect astrocytes remains limited. This thesis attempts to improve our understanding of how the activity of the nsP3 MD drives CHIKV preferential infection of astrocytes and investigate new methods to characterize the cell-
type dependence of CHIKV neurovirulence. An important observation in this study was that cells infected with CHIKV with decreased ADPr binding and hydrolase activity in the nsP3 MD (G32S) displayed more dsRNA staining than the cells infected with WT or CHIKV
with only impaired hydrolase activity (Y114A) in earlier time point, which is opposite to what has been observed before. The current observation might be attributed to host cellular dsRNA rather than viral dsRNA staining, which itself warrants further investigation. Additionally, translational activity indicating the establishment of CHIKV replication complexes is elevated in astrocytes compared to neurons, implying that the preference of CHIKV for infecting astrocytes is directly related to the ability to quickly translate viral RNA
CHARACTERIZING STRUCTURES PRESENT AT THE NK CELL IMMUNE SYNAPSE WITH QUANTITATIVE MICROSCOPY
Get PDFNatural killer (NK) cells are essential members of the innate immune system as
they aid in clearance of virally-infected and cancerous cells. NK cells have a cytotoxic
function that involves the formation of a synapse with a target cell, however the role of
endocytic proteins, which are known for their strong presence at cell membranes, have
been largely understudied in this context. Clathrin forms important structures at the cell
membrane and has been shown to have unique functions in B and T lymphocytes, but
the field has not yet explored this in NK cell synapses. While sophisticated microscopy
methods have been implemented to investigate the NK immune synapse in the
literature, these have rarely been applied to the study of endocytic proteins in real time.
Using light and electron microscopy techniques, this thesis aims to characterize clathrin
and dynamin structures at activated NK cell immune synapses with quantitative
methods. This work sets up future experiments and applications for correlative light
electron microscopy. Data produced by this thesis determined that clathrin and
dynamin are recruited to the NK immune synapse upon activation. With this, clathrin
forms differential structures at certain synaptic signaling domains. Dynamin-2 has a role
in lytic granule recruitment to the activated synapse, an essential step in NK cell
cytotoxic function. This work begins to describe a unique role for endocytic proteins in
synapse structure formation that has yet to be explored
From Chaos to Order: Assessing the Foundations of Emergent Gravity
No full textThe exploration of the relationship between gravity and thermodynamics has provided valuable insights into the search for a unified framework that reconciles gravity with quantum mechanics. In the 1970s, Jacob Bekenstein and Stephen Hawking established the thermodynamic nature of black holes, demonstrating that they possess entropy and thermal characteristics. Building on this foundation, Ted Jacobson’s work in the 1990s revealed a direct connection between the Einstein Field Equations and the thermodynamics of local causal horizons, hinting at a deeper link between gravity and thermodynamics. In the 2000s, Erik Verlinde introduced an entropic perspective on gravity, proposing that gravity arises as an entropic force driven by changes in information entropy. These developments suggest that gravity may fundamentally emerge from thermodynamic principles, prompting an inquiry into the nature of gravitational phenomena and their potential unification with quantum mechanics.
This dissertation explores the philosophical implications of emergence and reduction in thermodynamic approaches to quantum gravity. The historical development of emergent gravity is examined, tracing mechanistic explanations of gravity back to Isaac Newton. The validity of black hole thermodynamics is also assessed, countering recent philosophical skepticism by demonstrating that considerations of semiclassical quantum effects, such as the Unruh and Hawking effects, render such skepticism unfounded. Furthermore, the concepts of emergence and reduction are analyzed within the framework of Thermodynamic Gravity, revealing that it presents a case of emergence that remains consistent with reduction. Additionally, the principle of holography is explored in the context of Holographic Gravity, where I argue that gravity emerges as a manifestation of multi-scale entanglement relations
Optimization Strategies for Efficient Robotic Automation in Industrial Manufacturing Processes
Get PDFThe deployment of Automated Guided Vehicles (AGVs) and Autonomous Mobile Robots (AMRs) plays a key role in modernizing industrial operations by enhancing efficiency, precision, and adaptability. This study explores strategies for optimizing AGV and AMR deployment across diverse factory environments, focusing on system integration, workflow automation, and long-term scalability.
Case studies from three manufacturing facilities demonstrate practical implementation: (1) A large-scale AMR system at Grundfos in Houston, Texas, transported heavy-duty racks, addressing complex layouts and dynamic workflows. (2) A lightweight AMR at TE Manufacturing in Greensboro, North Carolina, efficiently handled small racks for compact and high-speed operations. (3) An autonomous forklift AMR at Pentair in North Carolina optimized logistics by managing heavy loads safely and precisely.
Key challenges include factory layout adaptation, seamless integration, fleet management, and real-time communication. Advanced technologies such as adaptive path planning, sensor fusion, and cloud-based monitoring enhance performance. This research provides practical recommendations, highlighting the potential of autonomous robotics to optimize manufacturing, reduce downtime, and improve efficiency, contributing to smart and sustainable industrial automation
PREDICTING RISK OF MAJOR ADVERSE KIDNEY EVENTS IN AFRICAN AMERICANS WITH DIABETES AND HYPERTENSION
No full textBackground: Chronic kidney disease (CKD) disproportionately affects racially minoritized populations in the US, with diabetes and hypertension as leading risk factors. This dissertation examines racial disparities in kidney function decline and major adverse kidney events (MAKE), focusing on the influence of demographic, clinical, and social determinants of health (SDOH). Methods: Data were sourced from the CURE-CKD Registry, comprising demographic, clinical, and SDOH information from a large Western US health system. Separate analyses were conducted for patients with diabetes and hypertension and included patients whose race is reported in the electronic health record (EHR) as non-Hispanic Black (NHB) or non-Hispanic, White (NHW). Unadjusted and adjusted Cox proportional hazards (CPH) models estimated the effect of race on MAKE. For each cohort, the predictive performance of the CPH model was compared to that of machine learning (ML) models. Shapley Additive Explanation (SHAP) values were used from the ML model with the highest average precision to assess each variable’s contribution to the model. Results: In both the diabetes (N= 375,605) and hypertension (N=710,768) cohorts, NHB patients had an increased risk of experiencing a MAKE outcome compared to NHW patients after adjustment for covariates. Insurance type, hospitalizations, and SDOH variables were significant predictors of MAKE in both cohorts. XGBoost ML models predicted the MAKE outcome with performance comparable to CPH models. The feature importance of ML models was generally consistent with the strongest predictors of the CPH models. Race was not identified as an influential feature to predict the risk of MAKE in either patient population. Conclusion: This study highlights significant racial disparities in MAKE and the value of incorporating SDOH into predictive modeling and interventions to mitigate adverse kidney outcomes
Artificial Intelligence-Based Clinical Decision Support in Cardiovascular Diseases
No full textSudden cardiac death (SCD) is a leading cause of mortality worldwide, with complex etiologies and risk factors that pose significant challenges for accurate risk stratification and timely intervention. Current clinical guidelines for SCD risk assessment in non-ischemic cardiomyopathies, such as hypertrophic cardiomyopathy (HCM) and cardiac sarcoidosis (CS), is suboptimal in capturing the multifaceted nature of the disease. This thesis aims to address these limitations by developing and validating novel artificial intelligence (AI) models for improved SCD risk prediction and clinical decision support in cardiovascular diseases.
The research presented in this thesis encompasses three main studies. The first study introduces a deep learning model for accurate detection and classification of heart rhythm abnormalities from 12-lead ECG data. The model architecture combines convolutional and recurrent neural networks to learn local and global features from ECG signals. The impact of lead selection on model generalizability is investigated, revealing that an optimal subset of ECG leads can improve performance and reduce overfitting.
The second and third studies focus on the development of multimodal AI models for SCD risk prediction in HCM and CS, respectively. These models integrate data from cardiac imaging and electronic health records to provide comprehensive risk assessments. The performance of the AI models is compared against existing clinical risk scores and guidelines, demonstrating significant improvements in predictive accuracy and risk stratification. Interpretability techniques are employed to identify the key clinical and imaging features driving the models' predictions, enhancing their transparency and potential for clinical adoption.
The results of these studies highlight the immense potential of AI in transforming clinical decision support for cardiovascular diseases. By leveraging advanced AI methods and diverse data sources, the developed models offer more accurate, personalized, and interpretable risk assessments compared to traditional approaches. The insights gained from this thesis lay the foundation for future research and clinical translation of AI-based tools in cardiovascular medicine, paving the way for improved patient outcomes and more efficient healthcare delivery
Efficient and Scalable Generative Model Control for High-Quality Multimodal Synthesis
No full textGenerative models, such as diffusion models and generative adversarial networks, have recently transformed foundational image synthesis tasks. However, challenges remain in designing generative models that are best suited for real-world applications and in improving their generalization capabilities across downstream tasks. This thesis addresses these challenges through comprehensive theoretical analyses and empirical experiments, with a focus on controllable generation tasks like text-to-image synthesis.
First, the thesis investigates the scaling properties of latent diffusion models, widely used in text-to-image generation. It introduces inference scaling laws related to this task and its downstream applications, revealing the surprising superiority of small models over large models when leveraging increased inference compute.
Next, it presents a novel scalable video diffusion model capable of generating continuous scenes from pure noise, extending generative capabilities from static images to dynamic, expressive videos. To further reduce the complexity of designing modality-specific models, the thesis proposes a versatile field diffusion model that can seamlessly handle various modalities, including image, video, 3D, and game simulation.
Additionally, the thesis introduces an efficient diffusion distillation technique that achieves comparable visual quality while reducing computational cost by 99%, significantly enhancing the sampling efficiency of generative models.
Building on these advancements, this thesis applies realism priors derived from generative models to three real-world image processing tasks: turbulence removal, shadow removal, and single-image super-resolution. These approaches consistently achieve state-of-the-art performance, surpass traditional regression-based methods, and deliver results with enhanced realism. Finally, the thesis demonstrates the applications of rectification in high-level vision tasks, including denoised image segmentation and super-resolved image re-identification
Algorithms for Quantitative Imaging with Advanced Cone-Beam Computed Tomography Configurations
Get PDFCone-Beam Computed Tomography (CBCT) is an emerging technology that offers several potential advantages over conventional multi-detector CT, including superior spatial resolution, point-of-care accessibility, and flexible mechanical designs enabling application-specific scanner configurations. When combined with the Dual-Energy (DE) acquisition technique, CBCT has the potential to quantify tissue compositions in the imaging target, such as bone mineral density (BMD) and oedema. However, quantitative accuracy of DE CBCT has generally been inferior to that of DE CT due to significant non-idealities in the imaging chain – e.g., elevated fraction of x-ray scatter reaching the detector – limiting its application in clinical settings. In this work, we bridge this gap through the development of advanced algorithms and experimental feasibility studies.
This work begins by a systematic investigation into the impact of x-ray scatter on the quantitative accuracy of DE cone-beam imaging. Simulations and physical experiments were performed for a representative DE task involving areal BMD quantification in the projection domain. Results show that material decomposition accuracy is highly sensitive to small scatter-induced biases, especially for DE protocols with poor spectral separation.
This finding motivated the development of a high-fidelity model-based framework for addressing these projection-domain non-idealities to obtain accurate volumetric DE decompositions from CBCT acquisitions. The proposed framework utilizes the inherent material discrimination ability of the DE data to facilitate the Monte Carlo-based scatter correction, with the addition of detector lag and glare pre-corrections. We demonstrated that this method improved the quantitative accuracy of wide-beam DE CBCT to levels comparable with DE CT in a challenging fat-bone-water decomposition scenario.
Following projection artifact corrections, we developed two practical DE material decomposition algorithms aimed at improving the quantitative accuracy in the presence of metal implants (which typically cause significant artifacts in reconstructed images). These algorithms – which incorporate different types of prior information into the model-based decomposition framework – were validated in an advanced multi-source DE CBCT system configuration and achieved improved BMD quantifications in fractures with orthopedic fixators compared to the existing approaches.
Lastly, we employed deep learning (DL) techniques to address residual artifacts in the reconstructed CBCT and DE CBCT images. Although many existing DL models have been developed to address image artifacts, there is a lack of efficient methods to quantify the reliability of each DL inference on-the-fly. In this work, we proposed an analytical pipeline to quantify voxel-wise uncertainties, and adapted it to an image fusion approach to correct residual CBCT artifacts in real-time. We showed that this method achieved an effective reduction of metal-induced artifacts in intraoperative CBCT imaging scenarios with minimal runtime, potentially improving the efficiency of the surgical workflow. Translation of this approach to DE applications of CBCT is ongoing.
We envision that this work can provide significant advancement toward translating point-of-care CBCT systems to clinical imaging applications requiring accurate biomarker quantifications, enhancing both the diagnostic and therapeutic workflows