Construction of LSTM Model to Assess Fall Risk Using Gait Conditions: Generation of Gait Data with Orthoses for Compensation

Abstract

This study proposes a Long-Short Term Memory (LSTM) model to assess the risk of falling among the elderly considering daily gait conditions. In constructing such an LSTM model, it is extremely difficult to collect gait data (time-series data of triaxial acceleration and triaxial angular velocity) from elderly persons at high risk of falling. This study addresses this problem by generating gait data with high risk of falling by healthy adults wearing orthoses developed in a previous study to compensate for knee extensor strength. Using the generated gait data, we developed an LSTM model to assess the gait state with high risk of falling, as quantified by the Timed Up and Go test score. The results show that the gait condition with high risk of falling can be estimated with 100% accuracy when the input time-series length exceeds 3 steps of gait data. Finally, this paper describes the use of the LSTM model developed in this study for transfer learning and fine-tuning to establish a model for fall risk assessment based on the actual gait condition of elderly people.