Using smart garments to differentiate among normal and simulated abnormal gaits

Abstract

Detecting and assessing an individual's gait can be important for medical diagnostic purposes and for developing and guiding follow-on rehabilitation protocols. Thus, an accurate, objective gait classification system has the potential to facilitate earlier diagnosis and improved clinical decision-making. Systems using smart garments represent an emerging technology for physical activity assessment and that may be relevant for gait classification. The objective of this study was to assess the accuracy of one such system - comprised of commercial instrumented socks and a custom instrument shirt - for differentiating among normal gait and four distinct simulated gait abnormalities. Eleven participants completed an experiment in which they completed several gait trails on a single day. Gait types were classified using diverse modeling approaches (K-nearest neighbors, linear discriminant analyses, support vector machines, and artificial neural networks). High classification accuracy could be obtained, both when classification models were developed and tested using data from each participant separately and grouped together, particularly using the k-nearest neighbor method (>98% accuracy). Some gaits were more often "confused" with other gaits, especially when they shared underlying kinematic aspects. These results support the potential of using "smart" garments for detecting and identifying abnormal gaits, and for future implementation in diagnosis and rehabilitation.

Keywords: Gait classification; Simulated abnormal gait; Smart shirt; Smart socks; Smart textile systems.