Semi-automatic segmentation for 3D motion analysis of the tongue with dynamic MRI

Abstract

Dynamic MRI has been widely used to track the motion of the tongue and measure its internal deformation during speech and swallowing. Accurate segmentation of the tongue is a prerequisite step to define the target boundary and constrain the tracking to tissue points within the tongue. Segmentation of 2D slices or 3D volumes is challenging because of the large number of slices and time frames involved in the segmentation, as well as the incorporation of numerous local deformations that occur throughout the tongue during motion. In this paper, we propose a semi-automatic approach to segment 3D dynamic MRI of the tongue. The algorithm steps include seeding a few slices at one time frame, propagating seeds to the same slices at different time frames using deformable registration, and random walker segmentation based on these seed positions. This method was validated on the tongue of five normal subjects carrying out the same speech task with multi-slice 2D dynamic cine-MR images obtained at three orthogonal orientations and 26 time frames. The resulting semi-automatic segmentations of a total of 130 volumes showed an average dice similarity coefficient (DSC) score of 0.92 with less segmented volume variability between time frames than in manual segmentations.

Keywords: Deformable registration; Dynamic MRI; Motion; Random walker; Segmentation; Super-resolution reconstruction; Tongue.

Figure 1

Dynamic MRI-based tongue motion estimation workow

Figure 2

Cine-MR (top row) and horizontally (middle row) and vertically (bottom row) tagged-MR images at three orientations.

Figure 3

An example of RW segmentation of the tongue. (a) A sagittal image of the region where the tongue touches the soft palate showing very poor image contrast between these two structures. (b) A user-given seeds separating the tongue (red) and the background (green) including the soft palate. (c) RW segmentation of the tongue (red) and the background (green).

Figure 4

An example of temporal stack segmentation at sagittal orientation. (a) Seed propagation from time frame 13 (user-given seeds) to other selected time frames (4, 8, 18, 22 in our case) by 2D deformable registration. (b) RW segmentation of temporal stack images using the user-given and propagated seeds in (a). (c) The user-given seeds on time frame 13. (d) An example of successful seed propagation. (e) An example of unsuccessful seed propagation.

Figure 5

Example of extracted seeds from the temporal stack of image segmentations. Bottom images show the seeds for the tongue (red) and the background (green) at three orientations extracted from the segmented masks on the top.

Figure 6

Example of super-resolution volume segmentations. Seeds were provided at 8 slices (3 axial, 2 coronal, 3 sagittal slices) in the super-resolution volume to segment the tongue at each time frame. (a) User-given seeds at time frame 13, and the segmented 3D tongue from the super-resolution volume. (b) Automatically extracted seeds from the temporal stack segmentations at time frame 1, and the segmented 3D tongue from the super-resolution volume.

Figure 7

Example of different seeding patterns in the back of the tongue region on Subject 1 (S1). (Top) Example axial slices on which seeds were input. SA1 and SA2 seeded on the axial slice 56 and SA3 seeded on slice 55. (Middle) User-given seeds for the tongue (red) and the background (green). (Bottom) Segmented tongue masks. SA3 used smaller ROI than SA1 and SA2. The tongue seeds (red) extend more in SA3 than the other two, resulting in larger segmented tongue (especially on the back).

Figure 8

DSC and volume box plots for the repeated semi-automatic segmentations and the manual segmentations. The red central mark is the median, the edges of the box are the 25th and 75th percentiles, and the whiskers extend to the most extreme data points. (a) DSCs between the repeated semi-automatic segmentations and the manual segmentations. (b) DSCs between the semi-automatic segmentations T1 and the other 7 sets of semi-automatic segmentations (T2-T8). (c) Volumes of the repeated semi-automatic segmentations. (d) Volumes of the manual segmentations.

Figure 9

An example of 3D displacement vectors of the tongue computed by the semiautomatic segmentations and HARP-IDEA. Time frame 8 (left column) shows the tongue motion from “a” to “s”, and time frame 17 (right column) shows the motion from “a” to “k” sound during a speech task of “asouk”.