Speckle Tracking for recovery of Displacement and Velocity Information from Sequences of Ultrasound Images of the Tongue

Sourabh Bora, MS student of Professor Mathews Jacob

CSB 523

Abstract

This paper explores the use of deformable registration (speckle tracking) as a method for obtaining point correspondences on sequences of tongue images acquired via ultrasound in order to estimate tongue velocity and displacement. We modify the velocity field as smooth Bspline functions and estimate it from adjacent frames. This model enables us to accumulate the motion and hence calculate the displacement from the first frame to any frame in the image sequence. Specifically, it provides the displacement estimates on the curve, enabling us to introduce "virtual flesh point markers" on the tongue surface. We tested the algorithm on ultrasound image sequences that were taken during the production of domain initial vowels in two different prosodic domains - word initial and accentual phrase initial. The results demonstrate the utility of the algorithm in quantification of tongue motion in speech.