DataSheet1_Intelligent Fall-Risk Assessment Based on Gait Stability and Symmetry Among Older Adults Using Tri-Axial Accelerometry.docx

This study aimed to use the k-nearest neighbor (kNN) algorithm, which combines gait stability and symmetry derived from a normalized cross-correlation (NCC) analysis of acceleration signals from the bilateral ankles of older adults, to assess fall risk. Fifteen non-fallers and 12 recurrent fallers without clinically significant musculoskeletal and neurological diseases participated in the study. Sex, body mass index, previous falls, and the results of the 10 m walking test (10 MWT) were recorded. The acceleration of the five gait cycles from the midsection of each 10 MWT was used to calculate the unilateral NCC coefficients for gait stability and bilateral NCC coefficients for gait symmetry, and then kNN was applied for classifying non-fallers and recurrent fallers. The duration of the 10 MWT was longer among recurrent fallers than it was among non-fallers (p < 0.05). Since the gait signals were acquired from tri-axial accelerometry, the kNN F1 scores with the x-axis components were 92% for non-fallers and 89% for recurrent fallers, and the root sum of squares (RSS) of the signals was 95% for non-fallers and 94% for recurrent fallers. The kNN classification on gait stability and symmetry revealed good accuracy in terms of distinguishing non-fallers and recurrent fallers. Specifically, it was concluded that the RSS-based NCC coefficients can serve as effective gait features to assess the risk of falls.

本研究旨在采用k近邻（k-nearest neighbor，kNN）算法，结合通过对老年人群双侧踝关节加速度信号进行归一化互相关（normalized cross-correlation，NCC）分析所得的步态稳定性与对称性指标，评估跌倒风险。本研究共纳入15名非跌倒者与12名复发性跌倒者，所有受试者均无临床显著的肌肉骨骼及神经系统疾病。研究记录了受试者的性别、体重指数、既往跌倒史以及10米步行测试（10 m walking test，10 MWT）结果。提取每次10米步行测试中段的5个步态周期的加速度数据，分别计算用于评估步态稳定性的单侧NCC系数与用于评估步态对称性的双侧NCC系数，随后采用k近邻算法对非跌倒者与复发性跌倒者进行分类。复发性跌倒者的10米步行测试时长显著长于非跌倒者（p < 0.05）。由于步态信号通过三轴加速度计采集，基于x轴分量的k近邻F1分数为：非跌倒者92%，复发性跌倒者89%；基于信号均方根（root sum of squares，RSS）的分类准确率为：非跌倒者95%，复发性跌倒者94%。基于步态稳定性与对称性的k近邻分类在区分非跌倒者与复发性跌倒者时表现出良好的准确率。具体而言，本研究证实基于均方根的归一化互相关系数可作为评估跌倒风险的有效步态特征。