Western University Open Repository

Western University’s Open Repository collects, archives, preserves, and freely disseminates scholarly works by members of the Western University community.

The Open Repository is also home to Western’s digital theses and dissertations.

Communities in DSpace

Select a community to browse its collections.

Now showing 1 - 3 of 3

Faculty and Staff Works 11856
Works by any staff member, including faculty, librarians, archivists, and post-doctoral fellows.
Graduate Student Works 17475
Scholarly works by graduate students, including theses and dissertations.
Undergraduate Student Works 1376
Scholarly works by undergraduate students. Undergraduate students must have faculty support and approval to post their work.

Recent Submissions

Access status: Open Access ,
Block-Based Finite Element Head Modelling for High-Fidelity Strain Prediction and Multi-Resolution Traumatic Brain Injury Analysis
(The University of Western Ontario, 2026-06-01) Islam, Sakib Ul
Finite element (FE) head models enable estimation of full-field brain deformation. The predictive performance of these models is governed by interdependent factors such as mesh topology, element formulation, material constitutive law, and strain output interpretation. This thesis systematically investigates these factors through all-hexahedral mesh construction, multi-domain model validation, functional parcellation, large-cohort impact reconstruction, computational scalability analysis, and representing explicit cortical gyrification. The thesis established that mesh topology plays a critical role in local strain prediction. Under matched loading conditions, block-based hexahedral meshes produced smoother strain distributions and more stable numerical behavior than previously used octree-based meshes, motivating a fully block-based approach for all subsequent models. Then, a high-fidelity 50th-percentile male FE head model was developed using node-matched skull-cerebrospinal fluid (CSF)-brain interfaces and region-specific Ogden hyper-viscoelastic brain properties, validated against cadaveric intracranial pressure, brain-skull relative motion, and regional strain and strain-rate data. Furthermore, a parcellation pipeline was developed to map the Schaefer functional atlas onto the FE brain mesh, enabling strain to be quantified within functionally defined cortical regions and networks. Applied to 53 reconstructed National Football League (NFL) head impacts to predict mild traumatic brain injury (mTBI) and initial clinical symptoms, the results showed that prediction of the mTBI risk improved when representing detailed functional regions, with the 400-region model outperforming whole-brain metrics for injury classification, while axonal-tract features contributed additional injury-related signal for symptom-specific outcomes. The thesis also demonstrated that computational efficiency can be achieved without sacrificing strain fidelity through controlled mesh coarsening, element formulation selection, and material recalibration. Finally, explicit modeling of cortical gyrification revealed that sulcal geometry concentrates strain in sulcal depths, exposing localized hotspots that commonly used smooth-brain models fail to capture.
Access status: Open Access ,
Understanding Medical Complexity and Transition Experiences in Long-Term Care: A Mixed Methods Study
(The University of Western Ontario, 2026-06-05) Chalmers, Karli
As the population ages, long-term care (LTC) demand is rising. This convergent mixed methods study used the Geriatric 5Ms (Mind, Mobility, Medications, Multi-Complexity, and Matters Most) to characterize medical complexity upon admission to LTC and contextualized these needs through transition experiences. Ontario LTC admission assessments (n=53,940; 2022–2024) were integrated with interviews from 11 caregivers and two LTC residents. Quantitative analysis found that 64.7% of residents had moderate-to-severe cognitive impairment, 84.2% required extensive assistance with activities of daily living, 65.5% experienced polypharmacy, and 94.9% were frail. Qualitative themes highlighted a reactive system where fragmented care and misaligned crisis thresholds forced caregivers to manage escalating needs until reaching a breaking point. Thus, medical complexity is not only a prerequisite for LTC placement, but a byproduct of a crisis-driven system. These findings demonstrate the need for increased access to community-based supports and navigation resources to improve transitions and minimize complexity on admission.
Access status: Open Access ,
Multimodal Machine Learning Frameworks for Solubility and Multicomponent Solid-Form Prediction of Pharmaceuticals
(The University of Western Ontario, 2026-06-02) Ghanavati, Mohammad Amin
Multicomponent solid forms, particularly salts and cocrystals, have emerged as effective strategies for tuning the physicochemical properties of crystalline materials, especially in pharmaceutical applications. These approaches are widely used to address key limitations of active pharmaceutical ingredients (APIs), such as poor aqueous solubility, which affects nearly 90% of new chemical entities and a significant fraction of marketed drugs. Despite their advantages, the experimental identification of suitable coformers or counterions remains a labor-intensive and uncertain process due to the vast number of possible molecular combinations. This thesis investigates the application of multimodal machine learning frameworks to improve the efficiency, accuracy, and scalability of computational screening for solubility prediction and multicomponent solid-form discovery. The central premise is that no single molecular representation sufficiently captures the complexity of intermolecular interactions, and that integrating complementary representations enables more robust and transferable predictive models. First, predictive models for aqueous solubility were developed using four curated datasets (ESOL, AQUA, PHYS, and OCHEM) comprising 3,942 unique molecules. Multiple molecular representations—including electrostatic potential (ESP) maps derived from 3,942 density functional theory (DFT) calculations, molecular graphs, and physicochemical descriptors—were explored. Individual models based on graph neural networks and feature-based learning achieved strong performance, with the best model reaching R² = 0.918 and RMSE = 0.613. An ensemble framework further improved predictive accuracy and robustness, achieving RMSE = 0.865 on the Solubility Challenge 2019 dataset, outperforming 37 benchmark models (average RMSE ≈ 1.62). Second, the role of physics-based information in cocrystal prediction was evaluated using a dataset of 7,395 molecular pairs (14,790 DFT calculations). A hybrid model combining graph isomorphism networks with molecular descriptors demonstrated superior predictive performance, achieving balanced accuracy of 0.916, and AUC of 0.97, outperforming both DFT-driven deep learning models and empirical approaches. Notably, this model eliminates the need for computationally expensive quantum chemical calculations during inference, making it suitable for large-scale virtual screening. Third, a unified machine learning framework (DualNet) was developed to simultaneously predict salts, cocrystals, and physical mixtures using a dataset of 22,298 experimentally validated systems. The model achieved strong generalization performance with macro-averaged recall of 0.952 and F1-score of 0.940. This enables confidence-aware ranking of candidate systems and significantly improves decision-making compared to traditional rules such as pKₐ-based classification. Finally, this work addresses a critical limitation in cocrystal prediction: the systematic absence of negative data due to underreporting of failed experiments. A multimodal semi-supervised learning framework was introduced to identify high-confidence pseudo-negative samples from large unlabeled datasets. By leveraging agreement across multiple molecular representations, this approach improves dataset balance and enhances model generalization without increasing computational cost. The results demonstrate that data-centric strategies can substantially improve prediction reliability in chemically diverse and previously unseen systems. Overall, this thesis shows that integrating multiple molecular representations with advanced machine learning techniques leads to significant improvements in predictive accuracy, generalizability, and practical applicability. These contributions provide scalable and reliable tools for accelerating solid-form screening and support more efficient design of pharmaceutical materials in both research and industrial settings.
Access status: Embargo ,
Challenges in Awareness: Hypertension Awareness Amongst Reproductive-Aged Adults in the Democratic Republic of the Congo and the United Republic of Tanzania
(The University of Western Ontario, 2026-05-25) Peters, Mariah Abigail Elisabeth
Hypertension prevalence has been increasing amongst reproductive-aged adults in sub-Saharan Africa, yet there is concern that awareness is not rising with it. Awareness in this population has been relatively unexplored in the Democratic Republic of the Congo (DRC) and Tanzania. Grounded in the political ecology of health framework and the health belief model, this thesis uses multivariate logistic regression to examine concordance between measured and self-reported hypertension in the DRC, as well as to analyze determinants of hypertension awareness in Tanzania. These findings indicate that there is a gendered gap in awareness. In the DRC, women were more likely to self-report being hypertensive, even when not measured to be, while men were the opposite. In Tanzania, women have higher levels of awareness. This thesis shows that self-reported data should be utilized with caution and context-specific policy should be put in place to increase hypertension awareness.
Access status: Open Access ,
A climatology of the stratospheric temperature using purple crow raman lidar
(The University of Western Ontario, 2009) Iserheinrhein, Blessing
The Purple Crow Lidar (PCL) is located at the Delaware Observatory (42° 52 N, 81° 23* W, 225 m elevation above sea level) near the campus of The University of Western Ontario. It consists of large power-aperture product monostatic laser radar which can measure temperature in the stratosphere. Stratospheric temperature measurements obtained from Purple Crow Raman N2 vibrational measurements during the years 1999 to 2007 are used to form a climatology of the stratosphere. Comparisons are made between measurements by the PCL and simultaneous radiosondes measurements in Detroit and Buffalo, as well as with atmospheric models, to determine the accuracy of the PCL measurements. An agreement to +1 K during summer months and ±2.5 K during the winter months is found. Comparison between PCL measurements and the CIRA atmospheric model show agreement of ±5 K or less. This results show that the PCL temperatures are valid and can be used for scientific studies.