123129 research outputs found
Sort by
Exploring Endovascular Photo-Activated Ablation (EPA) for Downstaging Locally Advanced Pancreatic Cancer: A Proof-of-Concept Study in the Normal Porcine Model
Get PDFBackground: Pancreatic cancers can involve large blood vessels early, making complete resection technically challenging or impossible. A minimally invasive treatment that clears vessels from encasing tumours could potentially enable curative surgery. We hypothesise that Endovascular Photo-activated Ablation (EPA) of perivascular tumour tissue can create a necrotic zone free of viable tumour between cancer and blood vessels, through which the tumour could be resected. Methods: A dose escalation study was conducted in the normal porcine model (n = 7). Under general anaesthesia, the animals were given a photo-activated drug and photo-activation was provided by a prototype balloon catheter, positioned in a major blood vessel within the pancreas, under angiographic guidance. Contrast-enhanced CT scans were undertaken prior to and 1, 2, or 7 days following ablation. The animals were euthanised and the exposed tissue excised en bloc for histological examination. Results: Five animals were euthanised after 2 days. On post-mortem, the histology confirmed necrotic pancreas in the perivascular zone, which increased from zero to 15 mm around the treated vessel, for increasing drug doses. Treated arteries showed necrotising arteritis, without evidence of perforation or obstruction during the observation period, although one animal was euthanised at 1 day, due to technical endovascular device issues and obstruction. The lowest-dose animal euthanised at 7 days showed no lesions on pathology. Conclusions: These proof-of-concept results demonstrate that EPA can produce pancreatic perivascular necrosis in a large animal model. In a pancreatic cancer abutting a major blood vessel, this procedure may be able to create a zone free of viable tumour, potentially rendering these cancers operable, while preserving vessel integrity. These findings support further research activities towards clinical translation
CO electrolysers with 51% energy efficiency towards C2+ using porous separators
No full textThis version of the article has been accepted for publication, after peer review and is subject to Springer Nature’s AM terms of use, but is not the Version of Record and does not reflect post-acceptance improvements, or any corrections. The Version of Record is available online at: https://doi.org/10.1038/s41560-025-01846-1Electrochemical CO2 reduction can generate multi-carbon (C2+) products via a CO2-to-CO cascade followed by CO reduction (COR). However, COR energy efficiency remains below 40% due to sluggish ion transport within charge-selective membranes. Here we introduce an uncharged porous separator that enables facile transport of both ion types, reducing ohmic resistance and superconcentrating cations at the catalyst surface—lowering COR voltage by 150 mV at 200 mA cm−2. In previous electrolyser designs, porous separators were limited by cathode-to-anode H2 crossover; the low diffusivity of C2H4 and CO in water allows a separator three times thinner and 1.6 times more porous, markedly reducing overpotential. Operating at elevated temperatures with a nickel–iron-based anode further lowers voltage by 330 mV, leading to a full-cell voltage of 1.95 V at 200 mA cm−2 and an energy efficiency of 51% to C2+ products sustained over 250 h. The system also achieves a CO single-pass conversion of 97% and a C2H4 concentration of 87 wt% in the product gas stream.D.S. acknowledges support from the Canada Research Chairs Program. R.K.M. thanks the Natural Sciences and Engineering Research Council (NSERC), Hatch and the Government of Ontario for their support through graduate scholarships. This work is financially supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada and TOTAL SE
A Data-Efficient and Generalizable Framework for Learning the Kinematics of Robots
Get PDFThis thesis presents a transfer learning framework designed to enhance the data efficiency and generalizability of learning the kinematics of robots. The framework leverages pre-training on simulated datasets to significantly reduce the amount of real-world data required to achieve high accuracy in both forward kinematics and inverse kinematics modeling. The effectiveness of this approach was demonstrated through comprehensive evaluations on concentric tube continuum robots.
For forward kinematics, the framework achieved sub-millimeter accuracy using only 10% of the real-world data needed by a baseline model trained from scratch. For inverse kinematics, it reduced real-world data requirements by 33% while maintaining sub-10 degree accuracy in rotational joint space errors. These results highlight the framework’s capacity to effectively utilize simulated data, ensuring robust pre-trained models that require minimal fine-tuning with real-world data.
The framework also demonstrated strong generalizability, particularly in its application to different robot designs. For forward kinematics, the framework achieved comparable accuracy with a 75% reduction in real-world data, while for inverse kinematics, a 33% reduction was observed. These findings underscore the robustness of the transfer learning approach, which adapts well to variations in robot configurations with minimal additional data.
Beyond concentric tube continuum robots, the principles underlying this transfer learning framework have broader applicability across various robotic systems. By reducing the reliance on extensive real-world data, this approach accelerates the development of accurate and reliable robotic models, making advanced robotic technologies more accessible and flexible for real-world applications.M.A.S
Deep Learning Based Surrogate Model to Estimate Scour Around Slab-on-grade Foundations Subjected to Flooding Conditions
Get PDFResidential foundations in coastal areas are susceptible to loss of soil support due to erosion and scour during storm surge events. The current methods to address these issues are largely based on empirical evidence and lack sensitivity to site-specific conditions. Thus, a research study was designed and performed to develop a computationally efficient deep learning based surrogate model to estimate scour around slab-on-grade foundations on sandy soils over a wide range of flooding conditions. A fully coupled 3-D numerical model, calibrated using published laboratory test studies, was used to simulate scour around slab-on-grade foundations. This model was used in conjunction with a space-filling experimental design to generate sample points required to develop a surrogate model. Two deep neural network (DNN) based surrogate models were developed and tested for predicting scour for varying flow and soil conditions. The DNN-based surrogate model, augmented with modified loss function to penalize un-physical predictions, was shown to achieve the best performance among the studied methods. Monotonicity analysis on the model showed that flood velocity, duration, and grain size properties of soils were found to have a major influence on the predicted maximum scour depth as compared to flood depth. The surrogate model developed can serve as an aid in achieving improved storm surge hazard preparedness, as well as proactive planning for post-hazard recovery efforts in vulnerable coastal communities.The presentation of the authors' names and (or) special characters in the title of the pdf file of the accepted manuscript may differ slightly from what is displayed on the item page. The information in the pdf file of the accepted manuscript reflects the original submission by the author
Bayesian Optimization Algorithms for High-Dimensional Problems and Robust Design
Get PDFIn this thesis, we develop novel approaches to address some of the challenges associated with Bayesian optimization of high-dimensional functions and optimization under uncertainty. We propose two representation learning approaches that enable Bayesian optimization algorithms to be scaled to high dimensions. The first approach, referred to as supervised function decomposition, involves learning a latent space where the objective and constraint functions can be approximated using additive Gaussian process models. This allows us to decompose the original high-dimensional optimization problem into a set of low-dimensional subproblems that can be efficiently solved independently. The second approach, referred to as supervised dimensionality reduction, involves learning a low-dimensional linear/nonlinear manifold mapping where the objective and constraint functions can be approximated. This reduced-order representation enables us to efficiently solve the optimization problem in a low-dimensional subspace and then map the solutions back to the original high-dimensional space using a decoder. Numerical experiments on benchmark test functions and a shape optimization problem with 114 design variables demonstrate that the proposed approaches can significantly outperform existing methods in the literature.
Finally, we develop a Bayesian framework for solving robust optimization problems that arise in applications where the objective and constraint functions depend on uncertain parameters. The key challenge in this context is that we only have access to observations of the original objective and constraint functions, not the robustness metric that we seek to optimize and/or the modified constraints in the robust optimization problem statement. We show how Gaussian process models of robustness metrics can be estimated given Gaussian process models of the original objective and constraint functions as a function of the optimization variables and the uncertain parameters. This development enables us to formulate acquisition functions that can be used to optimize the robustness metric subject to appropriate constraints. We also carry out a theoretical analysis of the proposed approach to gain insights into conditions under which the cumulative regret can be bounded. Numerical studies are presented on a set of test problems to demonstrate that the proposed approach holds significant potential for solving optimization problems under uncertainty on a limited computational budget.Ph.D
In Silico Model Development and Optimization of Lung Cell Population Dynamics
Get PDFThis thesis introduces a novel methodology for developing and optimizing in silico models predicting the growth and differentiation of cell populations under various culture conditions. By systematically combining first principles with methods from experimental design, statistics, and optimization, we formulate, calibrate, select, and validate biologically informed mathematical models. To illustrate the utility of the proposed methodology, we developed a series of mathematical models of increasing complexity for airway tissue engineering. Specifically, we developed the first mathematical model for the population dynamics of BEAS-2B cells, a human bronchial epithelial cell line commonly used in respiratory research. The model predicts cell growth based on initial population and nutrient and waste concentrations. Using this model, we determined an optimal media refresh schedule that enhances growth while minimizing the consumption of costly growth media.
We further applied the proposed methodology to multicellular populations to develop a mathematical model of the differentiation of induced pluripotent stem cells (iPSCs) into definitive endoderm (DE); iPSCs are adult cells reprogrammed to stem cells, and DE is a germ layer giving rise to organs like the lungs, liver, and pancreas. This model suggests no transitory state during differentiation expresses the DE biomarkers CD117 and CD184, a finding corroborated by existing literature. Additionally, the model indicates an optimal differentiation period of 1.9 to 2.4 days and identifies that lower plating populations result in higher DE yield per input cell.
Furthermore, we extended the multicellular model to account for the biochemical environment to investigate the differentiation of anterior foregut endoderm (AFE) into lung progenitors (LPs); both cells are precursors to respiratory tissues. The model indicates that daily media refreshment significantly enhances LP yield, compared with no change, approximately doubling it. It also shows that the LP yield per input cell on day 15 can increase 26% at lower split ratios.
This thesis underscores the utility of computational tools in tissue engineering by offering a systematic approach to optimizing cell culture conditions and experimental protocols. This approach complements traditional in vitro, ex vivo, and in vivo techniques, reducing experimental costs and timelines and improving protocol efficiency, with broad applications in regenerative medicine and drug development.Ph.D
Physics-based and Data-driven Modeling of Electrically Conductive Polymer Nanocomposites
Get PDFPolymer nanocomposites reinforced with carbon nanotubes (CNTs) have excellent mechanical, electrical, and electromechanical properties. Since the electrical properties vary with the mechanical applied load, interest in conducting polymer nanocomposite has increased due to their potential applications in strain sensing and structural health monitoring (SHM). An in-depth understanding of the structure-property relations of polymer nanocomposites reinforced with CNTs is needed to develop novel multifunctional materials. In this research program, a physics-based data-driven modeling framework capable of predicting the electrical and piezoresistive properties of CNT/polymer nanocomposites is developed. First, a physics-based stochastic multiscale model is developed using Monte Carlo simulations and representative volume elements. The developed numerical model is used to investigate the influence of the nanoscale parameters of CNTs and the microstructure of the nanocomposite on the percolation threshold, macroscopic electrical conductivity, and piezoresistivity of the CNT/polymer nanocomposites. Next, the numerical results from the developed numerical model are used to create representative datasets to train various machine learning models for efficient prediction of the nanocomposites properties. The developed framework is then used in the quantitative exploration of the structure-property relations of CNT/polymer nanocomposites to improve and accelerate the design of these multifunctional materials. The models developed in this research serve as tools for better understanding the underlying mechanisms of electrical conductivity and piezoresistivity in CNT/polymer nanocomposites. Moreover, the approaches used to develop these models can offer guidelines for modeling other multifunctional nanocomposites with embedded nanofillers. The insights gained from this research could be applied to SHM systems in modern structures like wind turbines, as well as within the aerospace and automotive sectors.Ph.D
Movement behavior and activity budgets of female Canada lynx (Lynx canadensis) during the denning season
Get PDFObserving the behavior of a species during its offspring-rearing period can provide valuable insights that improve our understanding of the habitat conditions required to support successful reproduction. We used GPS location data collected during a period of high snowshoe hare abundance from 2018-2020 to examine den use and investigate the capacity of Canada lynx (Lynx canadensis Kerr 1792) mothers to access prey resources around their den sites throughout the stages of kit development. We tested the hypothesis that mothers would be heavily restricted in movement during the first 2 months after parturition, at which point their behavior would return to pre-parturient baseline. At this high-latitude study site (~67°N), parturition occurred approximately three weeks later compared to lynx in the south (~47°N). Additionally, home ranges and core areas of females were greatly reduced (3-11x) following parturition. These observations, along with an estimation of activity budgets, suggest that there are both spatial and temporal restrictions affecting movement away from den sites even after kits are greater than 2 months old, indicating that females are limited to accessing prey resources within a constrained area (i.e., ≤ 10 km2) while caring for dependent kits.The presentation of the authors' names and (or) special characters in the title of the pdf file of the accepted manuscript may differ slightly from what is displayed on the item page. The information in the pdf file of the accepted manuscript reflects the original submission by the author
Occurrence, Prediction and Photocatalytic Oxidation of Substituted Polycyclic Aromatic Compounds in Industrial Wastewater.
Get PDFPolycyclic aromatic compounds (PACs) are ubiquitous organic pollutants occurring in industrial wastewater, particularly from steel mills and petrochemical facilities. Recent efforts have improved the monitoring and removal of PACs, but substituted PACs, such as alkylated (APACs) and nitrogen-sulfur-oxygen heterocyclic PACs (HPACs), are now being reported as a greater concern. These substituted PACs not only exhibit higher toxicity but are also resistant to conventional biological and tertiary treatment processes currently employed in the wastewater treatment plants. Additionally, some studies suggest that existing treatment methods may even promote the substitution of parent PACs, especially in coking wastewater. This data is needed to examine the adequacy of existing guidelines and regulations.
This study examines the occurrence of 61 PACs, including 16 parent PACs, 31 APACs, and 14 HPACs, in wastewater and sludge samples from two steel mills and one petrochemical plant in Southern Ontario. Detectable quantities of substituted PACs were found in the industrial effluents discharged to the receiving water bodies. A mass balance model applied to the influent and effluent wastewater samples indicated potential transformation of parent PACs to heterocyclic derivatives in one out of three industrial facilities during treatment, however variability in characterization methods prevented their precise quantification. Biological treatment step in the petrochemical refinery was found to be led by biodegradation while biosorption dominated the removal of substituted PACs in the wastewater treatment plant of steel mill.
Machine learning models were developed using standard wastewater quality indicators like COD, DOC, TSS, and NH₃-T. Support vector machine regression (SVR) accurately predicted HPAC concentrations in effluent (R² = 0.83). Model analysis revealed that SVR model weighed the input variables contributing to the substitution of PACs highly, and thus predicted the occurrence of more toxic substituted PACs with better accuracy. Such a model may be applied for the real-time monitoring of PACs in coking wastewater.
Solar photocatalysis using immobilized Ag/AgBr/TiO2 was found to effectively treat PAC-contaminated wastewater. A novel spectro-kinetic model using electron paramagnetic resonance (EPR) spectroscopy was proposed, which takes into account the temporal rise and decay of hydroxyl (•OH) and phenyl radicals (•CR), generated from electron (e −) and hole (h+) charge transfer on the catalyst surface. Oxidation pathways for direct h+ and •OH attack on substituted PACs were modeled to predict the concentration of a spiked compound in coking wastewater. This mechanistic kinetic model could be useful in photocatalytic process modelling, and ultimately in designing and optimizing wastewater treatment systems targeting substituted PACs.Ph.D
Programmatic, Environmental, and Temporal Predictors of Violence, Overdose, and Self-Injury in Homeless Shelters in Toronto, Canada, 2012-2021
No full textPeople experiencing homelessness have high rates of violent victimization, overdose, suicidality, and non-suicidal self-injury. These health-related critical incidents contribute to high mortality rates among homeless populations, making their prevention a critical public health goal. The objective of this study was to identify trends and correlates of physical violence, overdose, and self-injury in homeless shelters in Toronto, Canada. A retrospective study was conducted using administrative data on health-related critical incidents in Toronto’s shelter system from 2012-2021. Log-linked negative binomial regression models were fitted to (a) predict physical violence, overdose, and self-injury incident counts during the study period by year, season, and pandemic onset, and (b) examine programmatic, environmental, and temporal correlates of these outcomes, with separate analyses performed for the pre-pandemic and pandemic periods. Shelter-based physical violence (incidence rate ratio [IRR]: 1.08; 95% confidence interval [CI]: 1.06-1.11), overdose (IRR: 1.20; CI: 1.13-1.26), and self-injury (IRR: 1.16; CI: 1.11-1.21) incidents increased over the study period, with the rates of the increases for violence (IRR: 1.10; CI: 1.00-1.20) and overdoses (IRR: 1.66; CI: 1.48-1.86) intensifying during the COVID-19 pandemic. Larger congregate shelters had higher rates of violence, overdose, and self-injury, whereas shelter-based hotels had lower incident rates. Critical incidents were also higher during the winter and there was an increased rate of overdoses during check week. The development of smaller shelters that offer greater privacy to service users warrants further examination to advance safety in shelter settings. Increasing access to naloxone and other substance use supports, especially during check week, is also recommended for reducing drug-related harms