Improving Remote Patient Monitoring Systems Using a Fog-based IoT Platform with Speech Recognition

TL;DR

This paper tackles privacy, data flow, and interactivity challenges in Remote Patient Monitoring (RPM) by introducing a fog-based IoT platform that offloads processing from the cloud to edge fog devices and enables speech-based patient interaction. The authors design a low-cost testbed with NVIDIA Jetson TX1 as the fog node, Raspberry Pi end devices, and a cloud server to evaluate latency, throughput, and speech-recognition accuracy, achieving a training accuracy of $95.31\%$ and a validation accuracy of $88.10\%$ with a 70-30 data split. Results show significantly lower latency and higher throughput for the fog architecture compared to a cloud-only setup, indicating improved real-time responsiveness and data handling for RPM. The platform also provides an interactive interface via speech recognition and controlled video surveillance, enhancing RPM usability while maintaining data privacy through adaptive filtering at the fog layer. Overall, the work demonstrates that fog-IoT can enhance RPM performance and reach while offering a scalable path toward broader deployment.

Abstract

Due to the recent shortage of resources in the healthcare industry, Remote Patient Monitoring (RPM) systems arose to establish a convenient alternative for accessing healthcare services remotely. However, as the usage of this system grows with the increase of patients and sensing devices, data and network management becomes an issue. As a result, wireless architecture challenges in patient privacy, data flow, and service interactability surface that need addressing. We propose a fog-based Internet of Things (IoT) platform to address these issues and reinforce the existing RPM system. The introduced platform can allocate resources to alleviate server overloading and provide an interactive means of monitoring patients through speech recognition. We designed a testbed to simulate and test the platform in terms of accuracy, latency, and throughput. The results show the platform's potential as a viable RPM system for sound-based healthcare services.

Figures (5)

• Figure 1: Network hierarchy of proposed platform.
• Figure 2: General design flow.
• Figure 3: Model accuracy under varying dataset splits.
• Figure 4: Average latency of Fog and Cloud setups within 30 iterations.
• Figure 5: Average packet count of Fog and Cloud setups within 1 minute of data transmission.