Towards fully instrumented and automated assessment of motor function tests

Quantitative assessment of mobility and motor function is critical to our understanding and treatment of musculoskeletal and neurological diseases. Instrumented tests augment traditional approaches by moving from a single, often subjective, performance metric to multiple objective measures. In this study, we investigated ways of automatically capturing motor performance by leveraging data from a network of six wearable sensors worn at five different locations by 17 healthy volunteers while performing a battery of motor function tests. We developed a framework to segment motor tasks, e.g. walking and standing up, from 3D acceleration and angular velocity data, and extracted features. Results were compared to clinical test scores and manual annotations of the data. For the best performing sensors, we achieved a rate of correct classification of 82 to 100% and mean temporal accuracy of 0.1 to 0.6 s. We provided guidelines on sensor placement to maximize accuracy of the motor assessment, and a better interpretation of the data using our unsupervised subject-specific approach.