Conformal Wearable Sensors for Distinguishing Circumduction for Hemiplegic Gait with Machine Learning

Abstract

Circumduction represents a compensatory gait strategy that can arise as a consequence of hemiparesis. A circumduction gait strategy laterally deviates from the sagittal plane. Hemiplegic gait has been successfully quantified in the context of wearable inertial sensor systems involving both local and global wireless connectivity. Additionally, machine learning has been successfully applied to distinguish the affected leg and unaffected leg during hemiplegic gait based on the inertial sensor signal, such as from the gyroscope, which is consolidated to a feature set. The confluence of wearable systems with inertial sensors and machine learning implicates the evolutionary pathway to highly advanced rehabilitation opportunities. The next evolutionary phase involves the introduction of conformal wearable inertial sensor systems with a profile comparable to a bandage that utilize a segmented wireless architecture. Local wireless connectivity enables the operation of the conformal wearable, and global wireless connectivity permits inertial sensor signal data transmission to a Cloud for data storage and subsequent post-processing. During the post-processing phase software automation consolidates the inertial signal data to a feature set for machine learning classification. Circumduction is quantified by mounting a conformal wearable inertial sensor system for the hemiplegic affected leg and unaffected leg. Using the relevant gyroscope signal a multilayer perceptron neural network attains considerable classification to differentiate between the hemiplegic affected leg and unaffected leg during hemiplegic gait with respect to the context of circumduction. The combination of the utilities presented by conformal wearables, the Cloud, and machine learning offer highly augmented clinical acuity for highly advanced rehabilitation.