Korean J Geriatr Gerontol.  2022 Jun;23(1):44-50. 10.15656/kjcg.2022.23.1.44.

Unsupervised Machine Learning Analysis on Muscle Activities in Elderly’s Gait and Proposal for a Rehabilitation Exercise Method in Geriatric Physical Education

• Lee J

• KMID: 2566372
• DOI: http://doi.org/10.15656/kjcg.2022.23.1.44

Abstract

Background
The purpose of this cross-sectional study is to analyze the gait cycle of the elderly by dividing the gait cycle into four phases and extracting the weighting of eight muscles activities in each phases.
Methods
Six elderly patients participated in this study and measured muscle activities via electromyography during ambulation. After 20 gait cycles was determined, considering the signal noise and artifact, the weighting of the muscle activities in the four phases of gait cycle were analyzed via an autoencoder, which was an unsupervised machine learning technique.
Results
As a result, the characteristics of the specific muscle group in each phase of the gait cycle were found and, based on these results, a principle of rehabilitation exercise for the elderly’s gait ability was proposed. Among the four phases of the gait cycle, one phase, which was characterized by the high weighting of the ankle dorsiflexor, the lowest correlation (r=0.097) compared with the other phases between participants in the Pearson correlation analysis.
Conclusion
The high variability of a specific gait phase was associated with the falling risk, I proposed the importance of open-kinetic chain exercise to improve the lower extremity muscle coordination. Future studies should be designed to examine the clinical effects of exercise applying the proposed principle and usefulness of unsupervised machine learning as an evaluation tool of the elderly’s gait ability.

Keyword

Adapted physical education; Muscle coordination; Rehabilitation exercise; The elderly; Unsupervised machine learning