Real-Time Intelligent Patient Monitoring: A Federated Learning and TinyML-Based Approach

This chapter presents a patient position detection system for hospital environments, integrating an embedded system and a tiny machine learning-based (TinyML-based) predictive model enhanced with federated learning (FL) to preserve data privacy. An inertial measurement unit (IMU) sensor collects acceleration, gyroscope, and magnetometer data, which are transmitted in real-time via Message Queuing Telemetry Transport (MQTT) network. These data are normalised, scaled, and processed to extract relevant features, enabling real-time posture classification. The selected model, trained with a decision tree algorithm, identifies critical surgical postures such as lateral decubitus, supine, prone, and fowler. The training dataset was collected in a controlled laboratory environment using data from ten healthy volunteers aged 18-40 years, comprising university students and laboratory staff without musculoskeletal conditions. These positions are essential in hospital and therapeutic contexts, ensuring timely interventions, proper breathing, circulation, reduced pressure on support areas, and minimising tension on peripheral nerves. The classification model demonstrates efficient implementation on embedded devices using TinyML, supported by Firestore integration and model assembly. However, continuous evaluation of data quality, computational efficiency, and system latency is crucial for accurate and reliable real-time monitoring. FL enables each device to refine the artificial intelligence (AI) model locally using patient-specific data, sharing only model updates with a central server to protect privacy. The trained model, exported in C, runs on embedded systems, reducing processing time, improving care quality, increasing the likelihood of timely interventions, and supporting healthcare personnel in hospitals and home settings. This approach ensures a scalable and secure solution for patient monitoring, enhancing healthcare outcomes while maintaining data confidentiality.