Patient Stress Estimation for Using Deep Learning with RRI Data Sensed by WBAN

Patient's stress or positiveness during rehabilitation play an important role in determining the outcome of rehabilitation. Estimate of patient's stress must be useful to motivate training of patients without feeling stress and to improve effect of the rehabilitation. It is known that stress is correlated with vital signal such as electrocardiogram(ECG), electro-encephalogram(EEG), heartbeat and respiration. Wireless Body Area Network(WBAN) can measure patient's vital signal in real time during rehabilitation. In this research, we estimate patient's stress with neural network(NN) from their R-R interval(RRI) data of ECG measured by WBAN. We perform preprocessing based on medical knowledge and transfer learning method to improve stress estimation and reduce calculation complexity. Preprocessing extract part of features of stress before training of neural network. Through these methods, feature extraction in the NN became efficient, which not only reduced calculation complexity but also improved training speed and accuracy.