21793 research outputs found
Sort by
Enhancing Visual Interpretability in Computer-Assisted Radiological Diagnosis: Deep Learning Approaches for Chest X-Ray Analysis
Get PDFThis thesis delves into the realm of interpretability in medical image processing, focusing on deep learning's role in enhancing the transparency and understandability of automated diagnostics in chest X-ray analysis. As deep learning models become increasingly integral to medical diagnostics, the imperative for these models to be interpretable has never been more pronounced. This work is anchored in two main studies that address the challenge of interpretability from distinct yet complementary perspectives. The first study scrutinizes the effectiveness of Gradient-weighted Class Activation Mapping (Grad-CAM) across various deep learning architectures, specifically evaluating its reliability in the context of pneumothorax diagnosis in chest X-ray images. Through a systematic analysis, this research reveals how different neural network architectures and depths influence the robustness and clarity of Grad-CAM visual explanations, providing valuable insights for selecting and designing interpretable deep learning models in medical imaging. Building on the foundational understanding of interpretability, the second study introduces a novel deep learning framework that enhances the synergy between disease diagnosis and the prediction of visual saliency maps in chest X-rays. This dual-encoder, multi-task UNet architecture, augmented by a multi-stage cooperative learning strategy, offers a sophisticated approach to interpretability. By aligning the model's attention with that of clinicians, the framework not only enhances diagnostic accuracy but also provides intuitive visual explanations that resonate with clinical expertise. Together, these studies contribute to the field of medical image processing by offering innovative approaches to improve the interpretability of deep learning models. The findings underscore the potential of interpretability-enhanced models to foster trust among medical practitioners, facilitate better clinical decision-making, and pave the way for the broader acceptance and integration of AI in healthcare diagnostics. The thesis concludes by synthesizing the insights gained from both projects and outlining prospective pathways for future research to further advance the interpretability and utility of AI in medical imaging
Enhancing Consumer Access to Warehouse Clubs Amidst the Retail Food Access Divide: Assessing the Impact of On-Premises and E-commerce Services
Get PDFThis study explores consumer access to warehouse clubs in regions characterized by a pronounced retail food access divide, specifically where there is a high ratio of convenience stores to supermarkets and grocery stores. Despite the cost-efficiency and broad product offerings of warehouse clubs, their limited store network and membership fees raise questions about their capacity to attract consumers in areas with substantial gaps in food access. This research focuses on three key aspects: the direct effect of the food access divide on consumer visits to warehouse clubs, the influence of on-premises services on this relationship, and the influence of e-commerce services on this relationship. The results reveal that a high ratio of convenience stores to supermarkets and grocery stores significantly reduces consumer foot traffic to warehouse clubs. However, on-premises services, especially healthcare, not only boost consumer visits to warehouse clubs but also mitigate the negative impact of limited access to healthy food options on these visits. Automotive services also contribute positively, albeit to a lesser extent. E-commerce services have a nuanced role: while both home delivery and omni-pickup options decrease foot traffic directly, delivery services moderate the negative impacts of a high retail food access divide by sustaining consumer foot traffic. Conversely, omni-pickup services exacerbate the negative effects by further reducing in-person visits. This research enriches the literature on the retail food access divide, spatial resilience of warehouse clubs, and retailers’ channel capabilities. By integrating on-premises and e-commerce services, warehouse clubs can enhance retail service equity across diverse regions, particularly improving consumer accessibility in areas with significant retail food disparities
Enhanced Colorectal Polyps Screening with Optimized Barlow Twins
Get PDFColorectal cancer, despite being a leading cause of cancer deaths, is also highly preventable through efficient and fast diagnosis and precancerous lesions removal. However, bottlenecks in patient screening schedules prevent proper access to rapid diagnosis and emphasize the urgent need for efficient methods, such as Deep Learning (DL) tools, to support pathologists. Nevertheless, DL models face significant challenges in computational pathology because of the gigapixel image size of whole-slide images and the scarcity of detailed annotated datasets. It is crucial to leverage self-supervised learning (SSL) methods to alleviate the burden and cost of data annotation. However, current research lacks methods to apply SSL frameworks to analyze pathology data effectively.
We introduce a novel Barlow Twins framework, enhanced with an optimized augmentation strategy for pathology data. We leverage the KGH dataset, a private repository of colorectal polyps. We then train a Swin Transformer to exploit its hierarchical structure, effectively capturing the multi-scale nature of pathology images. These innovations improved Accuracy and Area Under the Curve (AUC) on the KGH dataset and PCam dataset, a well-known challenging benchmark for classifying metastatic cancer in breast cancer patients’ lymph nodes.
Furthermore, we provide meaningful explainability insights into the performance of the different techniques. We propose a practical and impactful approach for integrating deep learning tools into pathologist clinical workflow.
In this thesis, we demonstrate that the proposed model, relatively new to computational pathology, achieves remarkable and explainable results on various cancer types when adapted to the specific pathology task
Towards Cardiopulmonary Resuscitation Automation using Machine Learning
No full textCardiopulmonary resuscitation (CPR) is a critical intervention aimed at restoring vital blood circulation and breathing in individuals experiencing cardiac arrest or respiratory failure, representing a crucial aspect of emergency medical care. Numerous biomedical signals are associated with CPR execution and monitoring, from initial out-of-hospital treatment to the hospital's intensive care unit (ICU). Machine learning (ML) can play a crucial role in automating the CPR process by utilizing signals to identify complex patterns and in decision-making. In this context, we explored the existing role of ML in CPR and analyzed current ML approaches for various CPR-related tasks in this thesis. Our review highlights research gaps and sets new directions for empirical studies, uncovering the unexplored potential of ML applications in CPR. Through the analysis, we identified that CPR signals often suffer from noise, complicating accurate clinical interpretation and decision-making. Conventional denoising methods using filters exhibit limitations in addressing the complex noise characteristics inherent in CPR signals. Although ML is known for handling complex data characteristics, a dedicated ML-based unsupervised approach for denoising CPR signals is still missing. To this end, we proposed a novel ML framework tailored for denoising biomedical signals during CPR. Utilizing a multi-modality approach, our framework leverages a dedicated ML algorithm for individual signals while concurrently denoising multiple signals through unsupervised ML approaches considering real-life scenarios. Our framework demonstrates significant noise removal and signal fidelity enhancements. Furthermore, our methodology preserves signal correlations, essential for downstream tasks. Finally, the proposed framework aims to improve CPR monitoring and decision-making, offering adaptability and extensibility to denoise a range of biomedical signals beyond CPR scenarios
Authentication Protocols for IoT Edge Computing
Get PDFThe proliferation of IoT has led to vast interconnectivity, generating massive data that exceeds the processing capabilities of IoT devices. Traditional IoT-cloud models, where devices offload computations to centralized cloud servers, are increasingly inadequate due to the expected surge in IoT devices, projected to surpass 75 billion by 2025. This growth intensifies cloud vulnerability to single points of failure and highlights the need for alternatives that meet QoS requirements like low latency and location awareness. The 3-tier IoT-edge-cloud architecture offers a solution by processing data at nearby edge nodes, improving location awareness, and mitigating single-point-of-failure. While this distributed approach meets QoS requirements, it introduces security challenges, such as offloading data to distributed edge nodes without prior registration. Additionally, an adversary can trace the edge node attached to the IoT device and compromise the privacy of an IoT device user. Moreover, many deployed IoT devices are vulnerable to hardware compromise and unauthorized access, raising significant privacy and security concerns that hinder the broader adoption of edge computing.
In this thesis, we address the above challenges by proposing efficient and secure authentication protocols for IoT applications in edge computing. Our proposed protocols include Symmetric Key Authentication with Forward Secrecy (SKAFS), Symmetric Key Inter-Cloud Authentication and Redeemable Micropayment Protocol (SKICAP), Mutual Authentication Privacy-Preserving Protocol with Forward Secrecy (MAPFS), and Conditional Privacy-Preserving Message Authentication for VANET Emergency Exchange (CP-MAVE). The proposed protocols utilize lightweight cryptographic primitives to realize efficient protocols for edge computing. Moreover, the proposed protocols fulfill the security requirements for IoT applications, such as IoT device anonymity, session unlinkability, and resilience to hardware compromise of IoT devices.
For our proposed protocols, we provided formal security analyses based on computationally hard problems. Furthermore, we evaluated their performance in terms of communication overhead and computational complexity and compared them with other closely related protocols. Finally, we implemented prototypes of our proposed protocols using socket programming, simulating the message flow between the protocol entities to calculate their end-to-end latency and confirm the efficiency of our proposed protocols.
The proliferation of IoT has led to vast interconnectivity, generating massive data that exceeds the processing capabilities of IoT devices. Traditional IoT-cloud models, where devices offload computations to centralized cloud servers, are increasingly inadequate due to the expected surge in IoT devices, projected to surpass 75 billion by 2025. This growth intensifies cloud vulnerability to single points of failure and highlights the need for alternatives that meet Quality of Service (QoS) requirements like low latency and location awareness. The 3-tier IoT-edge-cloud architecture offers a solution by processing data at nearby edge nodes, improving location awareness, and mitigating single-point-of-failure. While this distributed approach meets QoS requirements, it introduces security challenges, such as offloading data to distributed edge nodes without prior registration. Additionally, an adversary can trace the edge node attached to the IoT device and compromise the privacy of an IoT device user. Moreover, with 2.38 billion IoT devices vulnerable to hardware compromise and unauthorized access, raising significant privacy and security concerns that hinder the broader adoption of edge computing.
In this thesis, we address the above challenges by proposing efficient and secure authentication protocols for IoT applications in edge computing. Our proposed protocols include Symmetric Key Authentication with Forward Secrecy (SKAFS), Symmetric Key Inter-Cloud Authentication and Redeemable Micropayment Protocol (SKICAP), Mutual Authentication Privacy-Preserving Protocol with Forward Secrecy (MAPFS), and Conditional Privacy-Preserving Message Authentication for VANET Emergency Exchange (CP-MAVE). The proposed protocols utilize lightweight cryptographic primitives to realize efficient protocols for edge computing. Moreover, the proposed protocols fulfill the security requirements for IoT applications, such as IoT device anonymity, session unlinkability, and resilience to hardware compromise of IoT devices.
For our proposed protocols, we provided formal security analyses based on computationally hard problems. Furthermore, we evaluated their performance in terms of communication overhead and computational complexity and compared them with other closely related protocols. Finally, we implemented prototypes of our proposed protocols using socket programming, simulating the message flow between the protocol entities to calculate their end-to-end latency and confirm the efficiency of our proposed protocols
Rekindling Creativity and Preventing Burnout: A Heuristic Inquiry into the Impact of Poetry and Photography on Art Teachers' Well-Being
Get PDFBurnout among art teachers has become a significant concern, marked by emotional exhaustion, depersonalization, and diminished personal accomplishment. This thesis explores innovative strategies to prevent burnout and rekindle passion for art among educators by integrating heuristics, poetry, photography, and mindfulness into their daily practices. Utilizing a heuristic methodology, the study emphasizes self-inquiry and discovery, offering art educators a structured yet flexible approach to navigate professional challenges. The research draws on Julia Cameron's principles from The Artist's Way and Mihaly Csikszentmihalyi's concept of "flow" to foster creative resilience and optimal engagement. Additionally, it incorporates insights from Jon Kabat-Zinn's Mindfulness-Based Stress Reduction and Ellen Langer's mindful learning framework to enhance emotional regulation and well-being. Through a studio-based approach, this study documents the experiences of the researcher-participant, highlighting effective methods for integrating small artistic habits and mindfulness practices into daily routines. The findings aim to provide actionable strategies for sustaining artistic engagement and well-being, ultimately contributing to the discourse on teacher retention and the development of supportive educational environments.
Keywords: Burnout; Art teachers; Mindfulness; Creative resilience; Teacher retentio
Visual Comfort Control Strategy for an Advanced Fenestration System in an Office Space
Get PDFBecause of the current trends in buildings which favor a higher window to wall area ratio, the impact of fenestration on the indoor environment has become crucial. New technologies appearing in the market make windows an active tool in controlling visual and thermal comfort, while allowing for energy generation through renewable sources such as photovoltaics.
The goal of this study is to present a control strategy for an integrated venetian blinds system within a triple glazed window with bifacial silicon photovoltaic cells on the outer glazing to optimize occupant visual comfort within a one-person office space. To achieve this, four objectives were considered. First, a visual transmittance model was developed to mimic the real-life behavior of the window under clear and cloudy conditions. The model was then integrated into a control strategy that uses the fenestration as an active tool for ensuring optimal visual comfort for office related activities, while reducing the energy used for heating by controlling passive solar gains. The control strategy determines the optimal blind tilt angle at each time step based on the outdoor climate conditions and occupancy schedule of the space. Next, the model and control strategy output were validated using measured data from an outdoor test-room representing an office space in Montreal, Quebec. Finally, a sensitivity analysis was conducted to determine the impact of physical parameters of the indoor environment on the visual comfort of the occupants. The window to wall ratio, the reflectance of the surfaces, and the room geometry were analyzed through simulation.
Using the control strategy, the results show that under clear sky conditions, the space can be self-sufficient in terms of illuminance levels but deals with certain levels of glare throughout the day. However, even though glare is imperceptible throughout the occupancy period under cloudy sky conditions, the illuminance levels do not reach the required 300 lux threshold for 23% of the day, requiring the integration of an artificial lighting source to fill in the missing gap to achieve the needed levels. In terms of physical properties of the space, it was found that an increase in the room dimensions leads to a decrease in illuminance levels, while a decrease in window to wall ratio also has the same impact. The surfaces reflectance also affects visual comfort, since highly reflective surfaces increase the work plane illuminance compared to more opaque surfaces. Overall, the control strategy presented in this work can be scalable and applicable to any type of office space that uses a similar advanced fenestration system
Consumer Perceptions of Authenticity: Cultural Expertise and Cosmopolitanism
No full textMany products and services today are advertised as “authentic;” however, not all consumers are equally knowledgeable about what this means. Cultural products, including tequila, are marketed with culturally congruent cultural symbols to increase consumers' perception of authenticity. However, research suggests consumers vary in their level of cultural expertise based on either their cultural identification or cosmopolitanism, which may moderate this relationship. This paper addresses the gap in the literature by measuring participants' level of cultural knowledge by the number of cultural symbols they recognize as Mexican and examines whether a participant’s expected cultural expertise can explain how many cultural symbols participants recognize, which explains their increased accuracy in identifying authentic product labels. The results illustrate that the more cultural cues participants recognize, the better they are at correctly identifying objectively authentic product labels. Although this paper finds no grounds for cultural expertise to moderate this relationship, consumers vary in their knowledge of cultural symbols, which should be considered when marketing to multiple consumer segments. It is suggested that ethnic products be marketed to a broader consumer market ethically, and future research could explore whether interest in a particular culture can explain the increased level of cultural knowledge
From Cocoon to Community: The Role of Silk in Intercultural Learning, Research-Creation and Art Education
Get PDFCan silk inform us of the past, the present and a reimagined future? This thesis mends together familial histories, intercultural learning, and community-based art education through research-creation. The research addresses silk’s history, production as a collaborative and cultural material in two components, a body of work exhibited at the Centre Culturel Georges-Vainer in Montréal, and a community workshop. The exhibition examines my first research question, above, and the silk moth through weaving, embroidery, and paper works alongside the forgotten experiences of my great-grandmother and family in diaspora. The art-making workshop, guided by the same research question, brought together communities in Montréal through needlework, conversations on materials considerations, and a collective art piece displayed in the exhibition. The research uses the theoretical lens of intercultural arts education to investigate my second research question: how can silk threads, storytelling, and community projects like my workshop connect cultures together, specifically in a diverse and multicultural city like Montréal? By examining the stories woven into silk, I address my third research question, how can projects such as Silk Cocoon contribute to inclusive intercultural art education in community settings? I offer a glimpse into a reimagined future where the threads of the past and present converge, guided by a deeper understanding of our relationship with materials, community, heritage, and the self
A Multi-period supply planning problem for managing stochastic demand
Get PDFIn this thesis, we address the critical issue of drug shortages that has persisted during and after the Covid-19 pandemic. We develop a two-stage stochastic model aimed at understanding how a compounding pharmaceutical company can balance supplier selection, drug production, back orders, and inventory management when faced with uncertain demand.
To explore and resolve the application of two models under varying conditions, we introduce the Sample Average Approximation (SAA) and Lagrangean Decomposition methods. Additionally, to better simulate the uncertainties present in real-world scenarios, we employ Monte Carlo simulations to generate diverse cases, enabling the model to produce more reliable results