Estimation of Knee Joint Angle Using a Fabric-Based Strain Sensor and Machine Learning: A Preliminary Investigation

Abstract

Monitoring human knee kinematics has various health applications including in-home rehabilitation and longterm tracking of movements of people with knee disorders. We proposed a wearable system based on a stretchable strain sensor and investigated its feasibility to estimate the knee joint angle in tasks of walking and static knee flexion. A pilot study with six subjects was conducted in which participants were asked to walk and perform flexion exercises at multiple speeds. Two commonly used machine learning algorithms (neural network and random forest) were utilized to estimate the knee joint angle based on the strain sensor data. The performance of the proposed approach was assessed in an intra- and inter-subject evaluation. In the intra-subject evaluation., the average mean absolute error (MAE) in estimating the knee joint angle during the walking task and flexion exercises was 1.94 and 3.02 degrees., respectively., with a similar coefficient of determination R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.97. In the inter-subject evaluation., an average MAE of 4.14 degrees in the walking task and 6.97 degrees in the knee flexion exercises was achieved with a R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> of 0.90. Our results suggest the feasibility of our approach., which includes a fabric-based strain sensor and machine learning., to estimate the knee joint angle. In future, this method might be used in various applications including the fields of healthcare., virtual reality and robotics.