The recent technological advances in environmental monitoring coupled with the increasingly stringent effluent requirements being placed on waste treatment systems makes it vital to have a more complete understanding of how specific compounds in waste streams can impact wastewater treatment processes. Since activated sludge processes are recognized as one of the most often applied technologies in wastewater treatment, this study assesses the impacts of select toxic synthetic organic compounds (SOCs) on the activated sludge communities in two types of wastewater treatment reactors: a completely-mixed activated sludge reactor (CMAS) and a sequencing batch reactor (SBR). Commonly applied activated sludge monitoring parameters, such as solids analysis and substrate removal, are collected and correlated to the results of microscopic image analysis (IA) and direct gradient gel electrophoresis (DGGE) to monitor the response of the activated sludge communities to variations in operational conditions, including the incorporation of SOCs in the influent feed and varying the solids retention time. The results of this research indicate that the response of the activated community is highly dependent on the reactor configuration. The CMAS settling performance was more strongly correlated to the shape parameters, and the SBR settling performance was more strongly correlated to the size parameters, which is qualitatively supported by particle settling theory when considering that SBR flocs were found to be larger than the CMAS flocs. The SBR began to exhibit larger floc sizes and had a higher sludge volume index with the incorporation of SOCs, while the CMAS flocs became more spherical after SOCs were incorporated and exhibited more discrete settling. The molecular analysis results revealed that the community structure within the activated sludge system was transient in response to environmental variations. Banding patterns indicated that samples were more similar to other samples taken from the same reactor under the same operational conditions. Thus, as operational conditions were varied, sample banding patterns would also change, indicating transitions in the genetic composition, and ultimately the dominant species present, in response to environmental changes.

Military personnel often experience mild traumatic brain injury (mTBI) from exposure to improvised explosive devices (IEDs). Soldiers typically endure blast trauma from the IED pressure wave as well as blunt trauma from ensuing head impacts. Researchers have not reached a consensus on whether the biomechanical response from blunt or blast trauma plays a more dominant role in mTBI because the specific biomechanical sources of injury are often undetermined. Consequently, the goal of this dissertation was to conduct three separate studies in order to characterize the mechanical behavior of the brain after exposure to mTBI conditions. For Study 1, mild blunt and blast trauma were induced in Sprague-Dawley rats using a custom-built device. In-house diffusion tensor imaging (DTI) software was used to make 3-D reconstructions of white matter fiber tracts before and after injury (1, 4, and 7 days). Axonal integrity was characterized by examining the fiber count, fiber length, and fractional anisotropy (FA). In-house image analysis software also quantified the microstructural variations in Hematoxylin and Eosin (H&E) stained brain sections, where significant differences in parameters such as the area fraction (AF) and nearest neighbor distance (NND) correlated to voids that formed after water diffused extracellularly from axons. Study 2 employed a computational approach involving the development of a finite element (FE) model for the rat head followed by the simulation of blunt and blast trauma, respectively. FE parameters such as von Mises stress, pressure, and maximum principal strain were analyzed at various locations including the skull, cerebral cortex, corpus callosum, and hypothalamus to compare injury cases. Study 3 involved interruption mechanical testing of porcine brain, a suitable animal surrogate of human brain. Compression, tension, and shear experiments were performed at a strain rate of 0.1 s-1 to examine the differential mechanical response. Microstructural changes in H&E stained brain sections were analyzed with in-house image analysis software to quantify differences among stress states at strains of 0.15, 0.30, and 0.40. Studies 1 and 2 confirmed that the brain behaves differently in response to blunt and blast trauma, respectively, while Study 3 further demonstrated the stress state dependent behavior of brain tissue.

No description available.

Obtaining ground-truth label information from real-world data along with uncertainty quantification can be challenging or even infeasible. In the absence of labeled data for a certain task, unsupervised domain adaptation (UDA) techniques have shown great accomplishment by learning transferable knowledge from labeled source domain data and adapting it to unlabeled target domain data, yet uncertainties are still a big concern under domain shifts. Distributionally robust learning (DRL) is emerging as a high-potential technique for building reliable learning systems that are robust to distribution shifts. In this research, a distributionally robust unsupervised domain adaptation (DRUDA) method is proposed to enhance the machine learning model generalization ability under input space perturbations. The DRL-based UDA learning scheme is formulated as a min-max optimization problem by optimizing worst-case perturbations of the training source data. Our Wasserstein distributionally robust framework can reduce the shifts in the joint distributions across domains. The proposed DRUDA method has been tested on various benchmark datasets. In addition, a gradient mapping-guided explainable network (GMGENet) is proposed to analyze 3D medical images for extracapsular extension (ECE) identification. DRUDA-enhanced GMGENet is evaluated, and experimental results demonstrate that the proposed DRUDA improves transfer performance on target domains for the 3D image analysis task successfully. This research enhances the understanding of distributionally robust optimization in domain adaptation and is expected to advance the current unsupervised machine learning techniques.

Statistical modelling has become a ubiquitous tool for analysing of morphological variation of bone structures in medical images. For radiological images, the shape, relative pose between the bone structures and the intensity distribution are key features often modelled separately. A wide range of research has reported methods that incorporate these features as priors for machine learning purposes. Statistical shape, appearance (intensity profile in images) and pose models are popular priors to explain variability across a sample population of rigid structures. However, a principled and robust way to combine shape, pose and intensity features has been elusive for four main reasons: 1) heterogeneity of the data (data with linear and non-linear natural variation across features); 2) sub-optimal representation of three-dimensional Euclidean motion; 3) artificial discretization of the models; and 4) lack of an efficient transfer learning process to project observations into the latent space. This work proposes a novel statistical modelling framework for multiple bone structures. The framework provides a latent space embedding shape, pose and intensity in a continuous domain allowing for new approaches to skeletal joint analysis from medical images. First, a robust registration method for multi-volumetric shapes is described. Both sampling and parametric based registration algorithms are proposed, which allow the establishment of dense correspondence across volumetric shapes (such as tetrahedral meshes) while preserving the spatial relationship between them. Next, the framework for developing statistical shape-kinematics models from in-correspondence multi-volumetric shapes embedding image intensity distribution, is presented. The framework incorporates principal geodesic analysis and a non-linear metric for modelling the spatial orientation of the structures. More importantly, as all the features are in a joint statistical space and in a continuous domain; this permits on-demand marginalisation to a region or feature of interest without training separate models. Thereafter, an automated prediction of the structures in images is facilitated by a model-fitting method leveraging the models as priors in a Markov chain Monte Carlo approach. The framework is validated using controlled experimental data and the results demonstrate superior performance in comparison with state-of-the-art methods. Finally, the application of the framework for analysing computed tomography images is presented. The analyses include estimation of shape, kinematic and intensity profiles of bone structures in the shoulder and hip joints. For both these datasets, the framework is demonstrated for segmentation, registration and reconstruction, including the recovery of patient-specific intensity profile. The presented framework realises a new paradigm in modelling multi-object shape structures, allowing for probabilistic modelling of not only shape, but also relative pose and intensity as well as the correlations that exist between them. Future work will aim to optimise the framework for clinical use in medical image analysis.

京都大学 / 0048 / 新制・課程博士 / 博士(農学) / 甲第21842号 / 農博第2355号 / 新制||農||1069(附属図書館) / 学位論文||H31||N5214(農学部図書室) / 京都大学大学院農学研究科森林科学専攻 / (主査)教授 大澤 晃, 教授 神﨑 護, 教授 井鷺 裕司 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DFAM

京都大学 / 新制・課程博士 / 博士(農学) / 甲第24663号 / 農博第2546号 / 新制||農||1098(附属図書館) / 学位論文||R5||N5444(農学部図書室) / 京都大学大学院農学研究科森林科学専攻 / (主査)教授 杉山 淳司, 教授 藤井 義久, 教授 仲村 匡司 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DFAM

Laser Doppler velocimetry (LDV) measurements and hydrogen-bubble flow-visualization techniques were used to examine the near-wall flow structure of 2-D and 3-D turbulent boundary layers (TBLs) over a range of low Reynolds numbers. The goals of this research were (1) an increased understanding of the flow physics in the near wall region of turbulent boundary layers, (2) to observe and quantify differences between 2-D and 3-D TBL flow structures, and (3) to document Reynolds number effects for 3-D TBLs. An ultimate application of this work would be to improve turbulence modeling for 3-D flows. The LDV data have provided results detailing the turbulence structure of the 2-D and 3-D TBLs, as well as low uncertainty skin friction estimates. These results include mean Reynolds stress distributions, flow skewing results, and U and V spectra. Effects of Reynolds number for the 3-D flow were examined when possible. Comparison to results with the same 3-D flow geometry but at a significantly higher Reynolds number provided unique insight into the structure of 3-D TBLs. While the 3-D mean and fluctuating velocities were found to be highly dependent on Reynolds number, a previously defined shear stress parameter was discovered to be invariant with Reynolds number. The hydrogen-bubble technique was used as a flow-visualization tool to examine the near-wall flow structure of 2-D and 3-D TBLs. Both the quantitative and qualitative results displayed larger turbulent fluctuations with more highly concentrated vorticity regions for the 2-D flow. The 2-D low-speed streaky structures experienced greater interaction with the outer region high-momentum fluid than observed for the 3-D flow. The near-wall 3-D flow structures were generally more quiescent. Numerical parameters quantified the observed differences, and characterized the low-speed streak and high-speed sweep events. All observations indicated a more stable near-wall flow structure with less turbulent interactions occurring between the inner and log regions for a 3-D TBL. / Ph. D.

High-grade serous ovarian cancer (HGSOC) is the most aggressive subtype of ovarian cancer, and its heterogeneity poses a challenge for the discovery of reliable diagnostic biomarkers, therapeutic targets, and predicting treatment response, particularly to immunotherapy. The current standard diagnostic and treatment options are inadequate, resulting in late diagnosis and poor prognosis. To improve our understanding of the immunophenotype of tumours, potentially enhancing diagnostic and treatment capabilities, the aim of the present study was to develop a stringent workflow for studying the immune microenvironment of HGSOC tumours. We utilized publicly available single-cell RNA sequencing data and literature to identify genes enriched in certain cell types of HGSOC tumours, followed by validation using immunofluorescent-based multiplex protein profiling. A 9-plex immunofluorescence workflow was developed using the Opal™ system, and quantitative image analysis was performed to evaluate the expression of PD-L1, CD8A, FoxP3, CD163, KRT7, PDGFRB, and CD79A in large tissue sections of ovarian cancer. Each of these markers are specific to different cell types, and by staining the multiplex marker panel together with new markers with little or no literature linked to HGSOC we can gain novel insights on the immune microenvironment of HGSOC. In this project, for a proof of concept, we focused on two proteins; GZMK and SLAMF7. The optimized multiplex panel developed as part of this project will be used to identify cell-type-specific markers that may play a crucial role in the immune microenvironment of HGSOC, which could lead to better immunophenotype stratification of patients and a more optimal immunotherapy response. Moreover, the panel could also be used to study markers of less well-known immune cell types, further improving our understanding of HGSOC. Overall, this project has the potential to significantly contribute to the development of reliable diagnostic biomarkers and therapeutic targets for HGSOC, ultimately improving patient outcomes.

In food manufacturing, the quality control procedure is a critical activity that consists in organizing, measuring, tracking, and filing the conditions of the production process and the final product, with the goal of guaranteeing the designed quality standard. During the last 30 years, due to a mounting concern by both consumers and lawmakers, the definition of quality and the application of quality control improved drastically, and new methodologies have been developed to ensure better control of food production and to understand the effect of raw materials and the process condition on the final quality of the food product. This thesis discusses the approaches to quality control procedures in food manufacture, focusing on the relationship between the conditions of the process and the quality profile of the final product, testing in a real-case scenario of a complex production process advanced data analysis procedures. The statistical and analytical procedures proposed have been applied in a real case studio from Trentingrana cheese production, a dairy consortium in the northeast region of Italy producing a ripened semi-artisanal hard cheese under the Protected Denomination of Origin (PDO) of Grana Padano. The aim is developing tailored statistical procedures that infer the effect of the critical factors of production on quality properties of this PDO product considering its semi-artisanal production process and the presence of multiple confounding factors. The statistical analyses were applied to a dataset of measurements of physical, sensory, and chemical properties collected on cheese wheels sampled systematically to represent the variability of the production of the Trentingrana wheels over two years of production. In the first introductory chapter, after a review of the different definitions of quality, the most important quality parameters for a food product and the standard measurement techniques adopted in quality control are presented. Then, in the chapter 2, the standard procedures of data analysis are reviewed, as well as the new approaches derived from the context of the foodomic sciences and machine learning models for the analysis of quality control data in food manufacturing. Two implemented and tested practical statistical procedures in the context of the Trentingrana consortium are reported: the results are discussed according to the objectives of the quality control process, the type of data, and the organization of food production. In the first case, reported in chapter 3, Linear Mixed Model ANOVA Simultaneous Component Analysis (LMM-ASCA) was developed to investigate the effect of the dairy factory, the bimester of production, and the variability within a cheese wheel using colorimetric and textural measurements. In the second case, reported in chapter 4, a standard ASCA model with the addition of a blocking factor to include systematic error was developed to investigate the relationship between the dairy factory and bimester of production and the volatile organic compounds (VOCs) profile of Trentingrana cheese wheels. In addition, in chapter 5, an approach to relate physical measurements on Trentingrana samples with sensory evaluations of texture by a trained panel is presented. The objective of this procedure is to incorporate the quality control procedure information from different quality parameters. The development of the Partial Least Squares (PLS) predictive model, its validation, and the evaluation of its performances are discussed. In the last section (chapter 6), the development of an image analysis procedure to measure the visual quality of the rind thickness of cheese wheels is reported, comparing the performances of two different algorithms. The data analysis tools proposed in this thesis have been proved to be useful for exploring, inferring, and plotting the process quality properties and suitable for analyzing complex and unbalanced experimental designs. Furthermore, the data analysis procedures proposed improve quality control activity both at the process level and at the product level, increasing the information that is possible to extract from the measurement collected in a context where standard statistical approaches cannot infer significant information.