Fall Risk Assessment Using Hybrid Machine Learning and Deep Learning Approaches and a Novel Posturography

The purpose of this project is to combine a novel posturogrpahy based on HTC VIVE trackers and hybrid machine learning and deep learning algorithms to establish a set of simple, convenient and valid fall risk assessment tool. This observational and follow up study will community elderly aged over 60 years old. The investigators will collect demographic data, questionnaire surveys, traditional balance tests (Berg Balance scale, Timed-up-and-go, 30s-sit-to-stand, four-stage balance tests) and a tracker-based posturography to obtain the trunk stability parameters in different standing task. The fall risk will be classified according to self-reported falls in the past one year and verified in a 6-month follow up.

The investigators will evaluate the performance of different hybrid machine learning and deep learning algorithm to extract the important features of multiple posturographic parameters and select an optimal model. The investigators will use the receiver operating characteristic curve analysis to compute the sensitivity, specificity and accuracy of different algorithms for risk classification and also compare the performance with traditional balance assessment tools. The investigators will evaluate the correlation of these posturographic features and data obtained by other methods. Risk factors of previous falls and future falls will also analyzed.