Model-based identification of motion sensor placement for tracking retraction and elongation of the tongue

Abstract

Electromagnetic articulography (EMA) is designed to track facial and tongue movements. In practice, the EMA sensors for tracking the movement of the tongue's surface are placed heuristically. No recommendation exists. Within this paper, a model-based approach providing a mathematical analysis and a computational-based recommendation for the placement of sensors, which is based on the tongue's envelope of movement, is proposed. For this purpose, an anatomically detailed Finite Element (FE) model of the tongue has been employed to determine the envelope of motion for retraction and elongation using a forward simulation. Two optimality criteria have been proposed to identify a set of optimal sensor locations based on the pre-computed envelope of motion. The first one is based on the assumption that locations exhibiting large displacements contain the most information regarding the tongue's movement and are less susceptible to measurement errors. The second one selects sensors exhibiting each the largest displacements in the anterior-posterior, superior-inferior, medial-lateral and overall direction. The quality of the two optimality criteria is analysed based on their ability to deduce from the respective sensor locations the corresponding muscle activation parameters of the relevant muscle fibre groups during retraction and elongation by solving the corresponding inverse problem. For this purpose, a statistical analysis has been carried out, in which sensor locations for two different modes of deformation have been subjected to typical measurement errors. Then, for tongue retraction and elongation, the expectation value, the standard deviation, the averaged bias and the averaged coefficient of variation have been computed based on 41 different error-afflicted sensor locations. The results show that the first optimality criteria is superior to the second one and that the averaged bias and averaged coefficient of variation decrease when the number of sensors is increased from 2, 4 to 6 deployable sensors.