Internet of Things (IoT) devices are revolutionising patient monitoring and in-home healthcare solutions, particularly for the elderly and People Living with Dementia (PLWD).
IoT devices can include motion accelerometer sensors, and smart cameras for remote patient monitoring that can help to prolong independent living for the elderly and people living with dementia.
Traditionally, IoT devices send highly sensitive personal data to the cloud for machine learning (ML) model training, posing privacy and security concerns. The transition to edge-based IoT applications is a significant step forward, enhancing scalability, privacy, and customisation, but individual-level accuracy can vary due to patient-specific data and evolving behaviours. We emphasise the importance of continuous, localised, and personalised retraining of ML technologies to maintain effectiveness.
PRISM leverages real data from an extensive clinical study and injects anomalies at the network gateway to assess performance. Neural network-based models, deployed at the edge, offer privacy-preserving, dynamically configurable solutions.
Key contributions of the research include injecting realistic anomalies, demonstrating the influence of training window size on accuracy, emphasising the need for patient-specific data, and showcasing feasible edge training on devices like Raspberry Pi.
This pioneering work, led by Imperial College London’s Institute for Security Science and Technology (ISST), opens new avenues for secure, personalised healthcare IoT applications and emphasises the critical role of edge computing in shaping the future of healthcare technology. We made our code and anomaly dataset available for further exploration and development in this evolving field.
Learn more about the PRISM research project here: https://petras-iot.org/project/privacy-preserving-iot-security-management-prism/