Digital Shape Reconstruction of Vocal Tracts from MRI Images

We present a method for computing surface models from MRI images with specific design for vocal tract extraction. As the vocal tract consists of empty space with connection to the outside, it is difficult to segment with common methods based on consideration of intensity differences. In the proposed method, surfaces are extracted via three steps. First, void voxels in an image of a human head are extracted using a morphological closing operator. Then, the significant and unnecessary noise contained in the results is removed via connected-component labeling. Finally, a vocal tract mesh is formed by applying the Marching Cubes algorithm in the extracted region and used to provide masked images. As the method does not require user input, it is efficient for batch processing.