Semi-automatic modeling of tongue surfaces using volumetric structural MRI
Daniel Bone, Michael Proctor, Yoon Kim, Shrikanth Shri Narayanan · The Journal of the Acoustical Society of America · 2011
Although volumetric magnetic resonance imaging has proven to be a valuable tool in the study of consonant production (Narayanan etal., 1995; Kröger etal., 2000), its utility is limited by the difficulty and laboriousness of reliably extracting tissue boundaries from imaging data. Current methods typically involve manual segmentation of air-tissue boundaries (e.g., Birkholz, 2006). Conventional automated (Atkins, 1998) and semi-automated (Ashton etal., 1995) methods used for the segmentation of brain MRI datasets may not be directly applicable to lingual segmentation because they are designed to work with different anatomical features. We present a method for extracting tongue surfaces from hi-resolution volumetric MRI data with limited user intervention. For each vocal tract volume to be analyzed, a lingual bounding box and search seed was first specified by an expert user, and voxel intensity was normalized across the region of interest. Lingual surfaces were automatically identified using a multi-pass region-growing algorithm operating over coronal planes. Thresholding was performed asymmetrically to allow for differential detection of air, teeth, and palatal boundaries, in opposition to adjacent lingual tissue. Smoothed tongue surfaces were fit to the resulting volumes by incorporating prior knowledge of intrinsic lingual musculature. [Work supported by NIH.]