82837 research outputs found
Sort by
Co-Training for Radiology Report Analysis
No full textThe growing popularity of natural language processing (NLP) has led to its widespread application across various domains, including medicine. In particular, radiology report analysis has explored many NLP techniques to improve information extraction and patient monitoring. However, a major challenge that continues to limit the broader adoption of NLP in this field is data scarcity. The need for expert-annotated data, which is both time-consuming and expensive to obtain, remains a significant barrier.
In this thesis, we present a novel insight to help address the data scarcity issue. Our intuition is that the structure of a radiology report—specifically, the \textit{Findings} and \textit{Impression} sections—offers distinct yet complementary views of the underlying radiology scan. Building on this insight, we introduce a co-training framework where two machine learning models are independently trained on these sections. By sharing information between the models, our approach enhances performance while utilizing large amounts of unlabeled data in a semi-supervised learning setting. We evaluated our approach through experiments conducted as part of a public health surveillance study, and the results demonstrate that our co-training method significantly enhances performance by utilizing the dual views of the report. Moreover, it consistently outperforms traditional supervised and other semi-supervised approaches. In summary, our contributions are: 1) We propose a co-training approach to radiology report analysis, which allows us to take advantage of the internal structures of a radiology report and available large amount of unlabeled dataset to address the data sparsity issue. 2) We craft a custom radiology report analysis dataset, motivated by real-world applications. 3) We conduct comprehensive experiments and analysis on the dataset, demonstrating the effectiveness of our method
Insight into Volleyball Biomechanics: A Markerless Motion Capture Analysis of the Spike Approach
No full textIdentifying injury risk due to altered techniques or dysfunctional biomechanics is a crucial area of study. Volleyball, a dynamic sport with inherent injury risks, necessitates a thorough understanding of biomechanics for injury prevention. It has the second-highest injury rate amongst team sports at the Olympic Games. Of the abundance of injuries, 56% occur at the net, which includes spiking and blocking. As volleyball is a non-contact sport, these injuries are primarily intrinsically occurring, meaning they are due to the individual and no outside factors.
Research is needed to determine if high-level volleyball athletes' biomechanical adaptations and altered techniques increase their injury risk compared to their counterparts with no reported injuries. There is currently research into injury rates in volleyball athletes, but there is a lack of research identifying biomechanical causes for these injuries. This study aimed to investigate biomechanical disparities in athletes during the volleyball spike by utilizing markerless motion capture technology. A comparison of injured and non-injured athletes was conducted to contribute to understanding biomechanics in volleyball and provide insight into injury prevention.
This study employed markerless motion capture technology to investigate biomechanical disparities during the volleyball spike approaches. This was focused on evaluating the potential of markerless motion capture technology to assess injury prediction on the volleyball court providing coaches and athletes with real-time risk data. Varsity volleyball athletes were measured to determine characteristic differences causing injury risk. This study focused on lower extremity injuries, which comprise around 45% of recorded volleyball injuries. A sub-selection of participants wore smart socks to determine the kinetic factors impacting takeoff and landing to
provide an understanding of the forces at play. The methods utilized allowed for a kinematic and kinetic understanding during the spike approach.
Due to the small sample size, non-parametric tests were utilized for analysis. A Mann- Whitney U test was performed to compare lower extremity joint angles at landing. While a Spearman’s correlation was run to assess the relationship between the parameters. The results confirmed significant correlations between the lower extremity joints, which displays the interconnectedness of the human body and shows how dysfunction at one joint impacts the surrounding joints. A significant positive relationship between previous ankle injuries and ankle kinematics at landing was found.
The findings emphasize the importance of addressing asymmetries and dysfunctional movement patterns to mitigate injury risk. Markerless motion capture systems were found to be useful for a biomechanical field assessment for high-level volleyball athletes. The study allowed for an understanding of the biomechanics at play during the volleyball spike. Thus, this research concludes that kinematic differences exist between athletes with and without previous lower- extremity injuries and can be quantified and measured with motion capture technology
Extraction of Dill Seed Oil/Essential Oil and Cranberry Pomace Anthocyanins using Green Technologies and In Vitro Evaluation of their Antifungal Activity
Full text linkClimate change has altered weather patterns that favor the growth of fungi on crops, therefore interest in natural alternatives, including essential oils and phenolic compounds, has increased due to concerns on the effects of chemical fungicides on the environment and human health. This study evaluated the use of supercritical CO2 (SC-CO2) technology to extract essential oil from dill seeds (Anethum graveolens) and high-intensity ultrasound (HIUS) with natural deep eutectic solvents (NADES) to extract anthocyanins from cranberry (Vaccinium macrocarpon) pomace. Both essential oils and anthocyanins have demonstrated antioxidative and antimicrobial properties, with potential applications as natural fungicides.
Dill seeds essential oil is known for its antimicrobial compounds, carvone and limonene. The extraction was conducted at temperatures ranging from 40 to 60 °C, pressures from 150 to 350 bar, and scaled up 3- and 10-fold. Three kinetic models were used to fit the data. The maximum yield of 13.69% was obtained at 350 bar and 50 °C. Then, commercial dill essential oil and SC-CO2 extracts were encapsulated via ultrasonic emulsification, using pea and lentil protein isolates, to test their antifungal activity against Fusarium graminearum and Pyrenophora tritici-repentis, two wheat pathogens. Ultrasonication at a nominal power of 1200 W produced emulsions with smaller particle sizes than high-speed homogenization, enhancing emulsion stability and protein solubilization. Pea protein isolate demonstrated stability comparable to synthetic emulsifiers. Extracts with higher carvone and limonene concentrations showed greater antifungal efficacy, with F. graminearum exhibiting a minimum inhibitory concentration (MIC) of 1.00 µL/mL, while P. tritici-repentis had MIC of 0.25 µL/mL.
On the other hand, cranberry pomace, rich in anthocyanins and phenolics, was investigated for its potential antimicrobial effects. The research optimized extraction by varying HIUS power levels and extraction times. Results revealed that a mixture of choline chloride + glycerol resulted in total anthocyanin content of 0.77 mg Cy3GE/g, comparable to ethanol (0.86 mg Cy3GE/g). Additionally, a mixture of glucose and lactic acid exhibited comparable extraction performance to ethanol in terms of total phenolic content (21.9 mg GAE/g). The extract of anthocyanins using choline chloride + glycerol was evaluated as an antifungal agent against F. graminearum and P. tritici-repentis. Although no antifungal activity against the strains was observed, the mycelial appearance suggested that choline chloride + glycerol may inhibit the production of fungal metabolites involved in F. graminearum pathogenicity. Anthocyanins powder from bilberries (Vaccinium myrtillus) were assessed. At a concentration of 4.0 mg/mL of anthocyanin, F. graminearum and P. tritici-repentis showed inhibition rates of 23% and 55%, respectively. This thesis highlights the potential of dill oil/essential oil and cranberry anthocyanins, green extraction techniques, and natural emulsifiers in encapsulation to create sustainable, effective antifungal agents for cereals like wheat and to contribute to sustainable agriculture
Effects of resistant starch on gut health and growth in pigs
No full textStarch from cereal grains, pulse grains, and tubers is a major energy substrate in swine rations constituting up to 55% of the diet. Starch types that are high in amylose content constitute a source of resistant starch (RS). The current state of knowledge indicates that RS as a prebiotic can enhance gut health of weaned pigs; however, the potential mechanisms still require further investigation. The overall objective of the thesis was to discover the interaction between resistant starch and the digestion kinetics along the gastrointestinal tract, which may induce alteration to the metabolites, butyrate production pathway genes, gut immunity specifically mucosal barrier function and immunomodulation, and energy metabolism in pigs.
In study 1, seven ileal-cannulated barrows (initial BW 30 kg) were fed 6 diets containing 96% of test ingredient (faba bean, field pea, chick pea, barley, wheat, or corn) or a N-free diet in a 7 7 Latin square at 2.8 maintenance DE with each experimental period lasting 9 days. The apparent ileal digestibility (AID) of starch (P < 0.001) was lower in pulse grains than in cereals, while hindgut fermentability (ATTD – AID) of starch (P < 0.001) was greater in pulses, particularly faba bean and field pea diets (P < 0.05). The greater fermentation of pulses contributed to higher digestible energy (DE) values; however, variations in protein and fiber interactions with starch granules may influence overall nutrient digestibility.
In study 2, a total of 32 weaned pigs were randomly allocated to 1 of 4 diets containing purified starch containing 0, 20, 35, and 70% amylose for 21 d. Increasing dietary amylose quadratically decreased average daily gain (ADG) and average daily feed intake (ADFI) (P < 0.01), decreased ileal digestibility of starch and increased hindgut starch fermentation (P < 0.001), leading to higher short-chain fatty acid (SCFA) concentrations in the cecum and colon (P < 0.01). Butyrate production was enhanced by upregulation of butyrate-producing pathway genes (BUT, BUK) (P < 0.05), while increased SCFA was negatively correlated with pro-inflammatory cytokines, IL-6 and TNF-⍺ (P < 0.05). The 35% amylose diet upregulated mucosal expressions of tight junction proteins in ileum and proximal colon (P < 0.05), suggesting an improvement in gut barrier integrity. However, increasing amylose downregulated lipid metabolism genes (FAS, ACC) (P < 0.05) in adipose tissue, muscle, and liver, potentially affecting fat deposition and energy utilization.
The results demonstrate that starch digestibility varies by botanical source, with pulse grains exhibiting greater hindgut fermentation than cereals. While RS fermentation supports SCFA production and gut health, its lower energetic efficiency compared to glucose may limit growth potential. The findings suggest that optimizing dietary RS inclusion can balance gut health benefits with sustained growth performance in pigs. Further research is needed to determine the ideal RS level, grain source, and feeding duration for maximizing both gut health and metabolic efficiency in swine nutrition
Getting a Sense of the nɬeʔkepmxcín Evidential nukʷ: Sensory evidence, modality, and Maximise Presupposition
No full textThis thesis examines the nɬeʔkepmxcin sensory evidential nukʷ. The focus of the thesis is on documenting the behaviour of nukʷ and developing a semantic analysis of it. Evidentials are words that encode the source of evidence for an utterance (Aikhenvald 2004). I argue that nukʷ is an evidential that encodes non-visual sensory evidence and is compatible with both direct and indirect evidence. It occurs with a range of nonvisual sensory evidence, including gut feelings and internal sensations. Using standard diagnostics for modality, I argue that nukʷ is an epistemic, variable-force modal with an evidential presupposition. Building on Peterson’s (2010) analysis of variable-force modality, I capture the behaviour of nukw using Kratzerian tools for analyzing modality. I discuss how the analysis predicts the relative distribution of nukw and the inferential evidential nke, which have similar evidence requirements. I propose that the choice between these evidentials is governed by Heim’s (1991) Maximise Presupposition when the evidence requirement of both evidentials is met. Additional findings include that nukʷ can give rise to mirativity, posing a problem for Rett and Murray’s (2013) account of mirative evidentials, and occurs in interrogatives, contra Littell and Mackie (2011), but only in non-flip interrogatives (cf. Bhadra 2020). Overall, this thesis contributes to our understanding of the types of evidence that can be encoded by evidentials, provides further evidence for overlap between the categories of modality and epistemic modality, and demonstrates how the choice between evidentials in a paradigm can be governed by informativity. It also raises questions for future research concerning evidentials in non-flip interrogatives and deriving the mirativity of evidentials
When Shape Matters: NS-SWORD Modelling of Neutron Star Shape Functions and the Impact on Solid Angles
No full textThe thermal emission from rapidly rotating neutron stars encodes information about their structure, which may be used to determine the underlying nuclear equation of state (EOS). Accurate modelling of this emission depends on the assumed surface geometry, typically parameterized by shape functions g(M,R,f) → R(θ) that take in the mass M, equatorial radius R and rotational frequency f of the neutron star and approximate the star’s oblate surface R(θ) as a function of colatitude θ. However, these shape functions may differ from the true surface computed using general relativistic equilibrium models for a given EOS. In this thesis, I present an investigation of what errors arise for shape functions in the literature, how they propagate to the solid angle subtended by the star as seen by a distant observer, and how shape errors affect the differential solid angle contribution from surface patches of the neutron star. To this end, I introduce and utilize NS-SWORD, a new code for modelling neutron star solid angles and surface fluxes based on the Oblate Schwarzschild (OS) approximation. Used in a pipeline with other codes, we can quantify the impact of R(θ) differences across a broad range of EOS, stellar parameters, and observer inclinations. Systematic comparisons are performed between the exact tabulated R(θ) calculated utilizing the Rotating Neutron Star Code (Stergioulas and Friedman (1995)) and all literature shape functions, including those from the works of AlGendy and Morsink (2014), Bauböck et al. (2013), Morsink et al. (2007), Silva et al. (2021), and the neural-network model of Papigkiotis et al. (2025). These comparisons reveal the different shapes of differences across θ and their magnitudes. They serve as the basis for the next set of comparisons between the modeled solid angle of a neutron star and the shape functions, where we build an intuition on how errors in the shape propagate into the errors in the solid angle. Finally, we explore how these errors propagate into the differential solid angle contribution of each patch of the surface of the neutron star, to understand how differences in R(θ) can affect localized solid angle contributions and to probe the problem of accurate global solid angles due to canceling local errors. The exploration of the differences in R(θ) reveals that no one shape function accurately models R(θ) without discernible issues across the entire parameter space of realistic neutron stars, although the shape functions of Papigkiotis et al. (2025) and Silva et al. (2021) are the better ones by the magnitude and comprehensibility of their errors, respectively. We identify the primary variable driving the error as the dimensionless spin parameter ϵ ≡ Ω^2R^3/M (in gravitational units, G = c = 1). These errors often produce percent-level discrepancies in the modeled solid angle, with the size and sign of the error varying with the spin parameter ϵ, compactness ζ ≡M/R (G=c=1), and observer inclination. While some functions achieve sub-percent agreement across much of the parameter space, no shape function is robustly accurate everywhere. Furthermore, “right answer for the wrong reasons” scenarios can occur where integrated solid angle errors are small but local compensating errors are significant. In our exploration of the differential solid angle contribution, we found that geometric effects are a dominant effect, biasing results for a given error in R(θ), but it is counteracted by gravitational self-lensing. We further find that the largest errors in differential solid angle occur close to the limb regions of the neutron star, whose contribution to the overall solid angle is small. However, these errors may prove significant if emission properties like temperature are stratified in latitude and location, as many properties are expected to be. These findings provide practical guidance for the selection and development of shape functions in neutron star modelling, and establish NS-SWORD as a tool for future studies of neutron star observables
Nicole Tegg - Abstract 14 - Innovate Conference 2025
No full textExercise-associated secondary amenorrhea results in estrogen deficiency, which may lead to dysfunction in estrogen’s normal cardioprotective pathways. Estrogen may be essential in a woman’s endothelial adaptations to exercise. The objective of this review was to assess the association between secondary amenorrhea in physically active women and cardiovascular disease (CVD) risk
Echoes of Home: The Profound Impact of Housing Insecurity on Women
No full textThe growing number of homeless people is a concerning issue, yet a large percentage of the homeless population - women - are often overlooked by policy actors. Women access services in different ways than men and are accounted for through the category of hidden homelessness. Even though women are present on the street visibly, they do not access emergency shelter services at the same rate as men because of fear and associated trauma. Drawing on interviews with 8 care workers and 6 homeless women to learn about their life experiences in navigating homeless care and homelessness, this thesis addresses the urgent need to amplify homeless women’s voices to understand their specific needs and experiences, which current housing solutions tend to neglect. Doing so helps explain the barriers women face in accessing services and the structural and social constraints that limit them. This thesis explores three living arrangements available to homeless women: (1) emergency shelters, (2) more stable housing provided following the Housing First Model, and (3) ultimately choosing to opt out of the formal system and instead to live in encampments. Community is at the core of this issue as women band together to move past fear and incidents of dehumanization to cope and heal from their traumatic pasts. Despite their agency and community, women face formidable structural barriers such as strict and discriminatory shelter rules, policy failures of the Housing First model, landlords’ control of the private housing market, and pandemic-related impacts on housing