7903 research outputs found
Sort by
AI Diagnostic Tool for COVID-19 Severity
Get PDFWith the advent of COVID-19 pandemic global, healthcare systems were challenged with resource constraints plaguing hospital environments around the world with consequences that presented weaknesses in our healthcare system. This demonstrated the importance of triage in unprecedented scenarios. This project presents an integrated AI-driven diagnostic tool to predict the severity of COVID-19 cases using length of stay as an analog. By using clinical data and radiographic chest images, this project integrates a XGBoost clinical data model with a Convolutional Neural Network (CNN) model utilizing chest CT images to create a comprehensive assessment of patient case severity through predicted length of hospital stay. Publicly available data was preprocessed and analyzed using techniques such as minimum Redundance Maximum Relevance (mRMR) feature selection and Spearman’s Rank correlation. The performance of the models was evaluated through multiple statistical metrics including Concordance Index(C-index), F1-score, Receiver Operating Characteristics (ROC), and Area Under the Curve (AUC). Through our project we demonstrated improved performance using the combined imaging and clinical data models with a F1-score of 0.860 and a sensitivity of 0.958. This demonstrates the model’s utility in making patient management more efficient and paving the way to more personalized medicine in clinical settings in future work. These models could be expanded to handle other potential lung diseases such as Tuberculosis and Pneumonia
Low-Power Association and Periodic Data Exchange Mechanisms for Wi-Fi IoT Devices
Get PDFThe adoption of Wi-Fi (IEEE 802.11) for Internet of Things (IoT) connectivity presents energy efficiency and reliability challenges, particularly for devices operating under stringent power constraints. Although recent enhancements such as Power Save Mode (PSM) and Target Wake Time (TWT) offer mechanisms for power savings, Wi-Fi was originally designed for high-throughput, continuously connected applications, and does not natively accommodate the low duty-cycle requirements typical of many IoT scenarios. This dissertation addresses the problem of how Wi-Fi-based IoT systems can achieve energy efficient and reliable connectivity while supporting periodic data exchange.
The investigation is structured through three empirical studies. The first examines the overhead associated with the association process across heterogeneous software and hardware configurations, identifying sources of inefficiency related to probing, key generation, and IP address assignment. Optimization techniques such as application processor-based key offloading and targeted Access Point (AP) probing are proposed and evaluated. The second study focuses on the impact of listen interval configurations on beacon reception, demonstrating that while longer intervals reduce wake-up frequency, they can increase total active time and energy consumption due to channel contention and timing uncertainties. The third study explores the interaction between TWT scheduling and Transmission Control Protocol (TCP) behavior, showing that misalignment between wake-up schedules and transport-layer expectations leads to premature retransmissions and degraded throughput. A cross-layer coordination mechanism is introduced to share TWT parameters with the TCP stack, enabling improved retransmission timeout estimation and congestion control.
The results of this work provide a comprehensive understanding of power consumption in Wi-Fi IoT devices and introduce practical methods to enhance energy efficiency and protocol reliability. The proposed techniques are validated across Real-Time Operating System (RTOS) and Linux-based platforms, demonstrating their applicability to a broad range of IoT deployments
