Prediction and Analysis of Fall Risk in the Elderly Using Multimodal Data
Jie Mao, Shigeho NODA, Zhe Sun, Ryutaro Himeno · Keisan Rikigaku Koenkai koen ronbunshu/Keisan Rikigaku Kouenkai kouen rombunshuu · 2024
Falls are a common problem among the elderly, often leading to severe physical injuries. It becomes more crucial to predict and prevent falls among the elderly. For this purpose, we aim to extract potential abnormal gait patterns and pre-fall indicators from multimodal data, which includes kinematic data, and gait data extracted from Inertial Measurement Units (IMUs). In this study, we first independently assess 31 elder people's Gait Risk Level using the Timed Up and Go test. Then we collected their gait data during Normal and Fast walking by IMUs. By calculating the gait trajectories from the IMU data and extracting the Average Stride Length and Minimum Foot Clearance (MFC), we used a classification model to classify the high, medium, and low risk based on Stride Length and MFC for both Normal and Fast-walking groups. The classification accuracy was 90% in the Normal speed group, increasing to 95% in the Fast-walking group. The result shows that the differences between the risk groups become more pronounced as walking speed increases. Particularly, there were significant differences in average stride length between the low risk group and the medium/high risk groups during both Normal and Fast walking group. The average values for both feet in the low risk group were closer and significantly greater than those in the medium and high risk groups. This result validates the effectiveness of our techniques and methods, providing a robust scientific basis for early diagnosis and prevention of fall risk in the elderly.