A Deep Learning and IoMT-Enabled Framework for Long-Term Human Respiratory Activity Assessment and Prediction with a Novel Flexible Wearable Sensor System

The respiratory monitoring research with the Internet of Medical Things (IoMT) can furnish essential early warnings and reliable safeguards for human health. Wearable respiratory monitoring devices exhibit advantages such as being compact, highly portable and capable of real-time monitoring. However, the currently available monitoring devices still suffer from unsatisfactory long-term wearing comfort, significant errors caused by environment, and signal distortions. In this work, we propose a novel wearable respiratory monitoring framework, termed DCAFP, which is focused on an innovative artificial muscle structure-based flexible wearable sensor system and several robust deep learning modules. This system can reconstruct and analyze the original respiratory data under diverse human activities and predict the long-term condition of human respiratory autonomous function. The framework first trains each teacher model separately, then the proposed multi-teacher knowledge distillation mechanism can be applied to continuously optimize the student model with better prediction performance. Our experiments show the mean square error (MSE) of 18.57 and mean absolute error (MAE) of 12.09, indicating the feasibility and effectiveness. This work has the full potential to shed light on the development of portable vital signs monitoring devices and long-term human respiratory function assessment and forecasting in related IoMT applications.