Tracing vocal fold vibrations using level set segmentation method

Abstract

High-speed digital imaging (HSDI) of the larynx can provide important information on the vocal fold kinematics. This information is useful and may provide a better understanding of the mechanism of phonation and assist clinical assessment of voice disorders. Automatic tracing of the vocal fold vibration is a key step in the kinematic analysis and for correlative characterization of vocal fold vibrations and voice quality in normal and diseased states. In this study, we introduce a new approach for image segmentation and automatic tracing of vocal fold motion that combines the level set method and motion cue. This approach is applied to videokymogram (VKG)-form images, which are obtained from a sequence of laryngeal images captured using the HSDI. To utilize the motion cue for a more effective level set based segmentation on the VKG, we first construct a so-called standard deviation (STD) image by mapping the pixel-based measure of temporal intensity dispersion from the initial HSDI sequence. The STD image maps the extent of vocal fold motion, and followed by threshold operation, a region of interest (ROI) that encloses vocal fold motion, or glottal region, is identified. The performance and effectiveness of our approach are evaluated by using clinical datasets representing both normal and pathological voice conditions.