Automated scoring of rehabilitative tests with singular spectrum analysis

In rehabilitation, continual assessment of those with disabilities is needed to determine the effectiveness of therapy and to prescribe the regimen and intensity of future treatment. Conducting assessments is challenging - there is a need to maintain objectivity and consistency across time. Also, repetitious tests can lull the assessor into lower levels of alertness. These motivate for automated scoring of rehabilitative tests. In this paper, we describe our work in automating the widely used and accepted Action Research Arm Test. We focus on the grasp subtest which employs a cube into which we embed sensors. Previously we have used live patient simulators and now the full set of patient trials have been completed. We employ Singular Spectrum Analysis on the signals, for which the resulting eigenvalues are then selected in a principled way to aid in signal filtering. The results show encouraging promise in our quest for automated scoring.