Skip to main content
. 2019 Jul 10;6(7):181982. doi: 10.1098/rsos.181982

Figure 2.

Figure 2.

The dataset included 400 swallowing cases accompanied by hyoid bone tracking annotations used to train the SRNN. In (a), the green areas indicating the hyoid bone are approximated for the reader’s convenience. In (b), the hyoid bone location in each frame was manually labelled by one experienced rater and the inter-rater reliability test was implemented with three other frame raters. The swallowing segmentation was also labelled by one rater and the inter-rater reliability test was implemented with another rater, as shown in (c).