A study on nursing assist system using deep reinforcement learning for prevention of patients falls

Abstract

We have been studying two main methods for preventing patient falls and reducing the workload of medical staff at medical and nursing care facilities, using artificial intelligence and robotics technologies. One is a realtime fall risk assessment of patients and their surrounding environment using deep learning. The other is a method to present recommendations and nursing assist to medical staff for optimal interventions for patients using deep reinforcement learning to reduce the risk. In this paper, we analyze the results of incorporating nursing theories into deep reinforcement learning, and introduce the effectiveness, limitations, and future prospects of the application. First, as applications problem of deep reinforcement learning, we apply representative nursing theories of Roy, Johnson, Orem, Henderson, Peplau, Orlando, Abdellah, King, Wiedenbach, Travelbee, Leininger, Newman, to achieve consistency with the medical field. Next,as an effect of their application, we show that it is possible to clarify what information from the clinical field should be used for the main design parameters of deep reinforcement learning, i.e., state, action, and reward. As a limitation of the application, the conventional training and evaluation by reproducing scenes in dynamic 3D simulations can be used to evaluate operational efficiency strategies and safety based on environmental sensing data and external feedback from patients on intervention. However, it is necessary to incorporate internal feedback into the evaluation of the validity of the actions of nursing assist robots and medical staff. Finally, the use of a virtual environment, such as a metaverse with avatars, will suggest future work to solve this problem