ETD

• 5G
• Artificial Intelligence (AI)
• Multi-Access Edge Comput- ing (MEC)
• SURGERY

The integration of advanced technologies such as 5G, Multi-Access Edge Computing (MEC), Artificial Intelligence (AI), and Federated Learning (FL) has emerged as a transformative approach to addressing critical challenges in resource-constrained healthcare environments. This thesis explores the design and experimentation of a comprehensive framework that combines these cutting-edge technologies to support real-time, accurate, and scalable decision-making in healthcare, particularly in surgical and diagnostic workflows.
Problem Statement
Traditional healthcare systems often rely on centralized cloud-based architectures, which are hindered by high latency, dependency on reliable internet connectivity, and significant energy consumption. These limitations are particularly pronounced in underserved and remote areas, where access to computational resources is minimal. The need for low-latency, energy-efficient, and scalable solutions to support AI-driven healthcare applications forms the cornerstone of this research.
Proposed Solution
This thesis presents an integrated framework leveraging:
• 5G Connectivity: A private 5G network ensures ultra-low latency, high-speed data transmission, and reliable communication between edge devices and MEC nodes.
• Multi-Access Edge Computing (MEC): MEC nodes host computationally intensive tasks, reducing dependence on centralized cloud infrastructure and enabling real-time data processing close to the source.
• AI Models: State-of-the-art convolutional neural networks (CNNs) such as YOLOv8n, MobileNetV3-Large, and DenseNet201 were optimized and deployed for real-time detection and classification tasks. Federated Learning (FL): A privacy-preserving framework for collaborative training across multiple institutions, addressing data scarcity and ensuring compliance with data privacy regulations.
Methodology
The research methodology involved a multi-phased approach:
• Deployment of a private 5G "5G-in-a-box" solution to establish high-speed connectivity.
• Implementation of MEC nodes for hosting AI inference tasks, significantly reducing end-to-end latency.
• Development and evaluation of AI models for surgical instrument detection, lung disease classification, and image-guided surgery applications.
• Performance benchmarking of MEC and cloud-based inference in terms of latency, energy consumption, and scalability.
• Utilization of Federated Transfer Learning (FTL) to optimize energy efficiency and minimize the environmental impact of AI model training.
Key Findings
The experimental results demonstrated the following:
• Latency Reduction: MEC infrastructure reduced end-to-end latency by over 70 times compared to cloud-based systems, enabling real-time applications such as augmented reality in surgery.
• Energy Efficiency: Edge devices powered by optimized AI models consumed as little as 1.2 W, aligning with green computing principles.
• Model Performance: Lightweight models like MobileNetV3-Large and YOLOv8n achieved high accuracy and low latency, making them ideal for edge deployment.
• Federated Learning Impact: FTL demonstrated a reduction in energy consumption compared to training models from scratch while preserving data privacy and achieving competitive accuracy.
Broader Implications
The findings underscore the transformative potential of combining 5G, MEC, AI, and FL in healthcare. By enabling real-time, scalable, and energy-efficient solutions, the framework bridges the gap between advanced AI technologies and their practical deployment in underserved areas. It also highlights the importance of privacy-preserving mechanisms and environmental sustainability in future healthcare innovations.
Conclusion
This thesis provides a robust foundation for integrating next-generation technologies into healthcare frameworks, demonstrating their feasibility and impact. The insights gained from this research contribute to advancing real-time, AI-driven healthcare solutions that are scalable, efficient, and accessible. Future work will explore hybrid approaches combining MEC and cloud resources to further enhance system performance and scalability while addressing emerging challenges in healthcare delivery.