A data-driven method to reduce excessive contact pressure of hand orthosis using a soft sensor skin

Abstract

Discomfort under customised hand orthosis has been commonly reported in clinics due to excessive contact pressures, leading to low patient adherence and decreased effectiveness of orthosis. However, the current orthosis adjustment by clinicians to reduce pressures based upon subjective feedback from patients is inefficient and prone to variability. Therefore, a quantitative method to guide orthosis adjustment was proposed here by developing a data-driven method. Firstly, Verbal Protocol Analysis was employed to convert the implicit process of orthosis customisation into working models of clinicians. Relevant data to inform a new solution development to reduce excessive contact pressure were extracted from the working models in terms of time consumption and iterations of tasks. Secondly, a new soft sensor skin with strategically placed sensing units to measure static contact pressures under hand orthoses was developed. Finite element simulations were conducted to reveal the required contact pressure range (0.02 – 0.078 MPa) and the distribution of relatively high pressures in 12 key areas. A new fabrication method was proposed to produce the sensor skin, which was then characterised and tested on the subject. The results show that the sensor unit has a pressure range from 0.01 MPa to 0.1 MPa with the maximum repeatability error of 6.4% at 0.014 MPa, and the maximum measurement error of 8.26% at 0.023 MPa. Thirdly, a new method was proposed to predict contact pressures associated with the moderate level of discomfort at critical spots under hand orthoses. 40 patients were recruited to collect contact pressures under two types of orthoses using the sensor skin, and their discomfort perceptions were measured with a categorical scale. Based on these data, artificial neural networks for five identified critical spots on the hand were built to predict pressure thresholds that clinicians can use to adjust orthoses, thus reducing excessive contact pressures. The neural networks show satisfactory prediction accuracy with R2 values over 0.89 of regression between network outputs and measurements. Collectively, this thesis proposed a novel method for clinicians to adjust orthoses quantitatively and reduce the need for subjective assessment for patients. It provided a platform to further investigate the pressure for patients with other conditions.