Yoshitomo Sakuma, Kento Takabayashi, Takumi Kobayashi, Chika Sugimoto and Ryuji Kohno
This paper proposes a learning scheme for the Recurrent Neural Network (RNN) modelto predict the anesthetic effect on a patient during surgery to support stable controlling anesthesia in clinical operation under lack of anesthesiologists. The goal of this study is to realizedependable control of total intravenous anesthesia (TIVA) to support anesthesiologists. To realizedependable control of anesthesia, it is necessary to satisfy various restrictions to guarantee safetyon anesthesia and the use of model predictive control (MPC) is effective in controlling anesthesiato meet those requirements. In the MPC, the stability of the prediction model is very importantto guarantee the control performance. Also,the RNN is useful for modeling nonlinear systems with time variations, such as patient drug response. Although stability of an existing learningscheme using the RNN has not been analyzed, the paper investigates the stability of the proposed learning scheme using the RNN by a manner of Lyapunov analysis.
Prediction Model, Total Intravenous Anesthesia, Machine Learning, Recurrent NeuralNetwork, Stability Analysis