Segmentação do tracto vocal a partir de estudos imagiológicos de ressonância magnética

Authors

Abstract

The main objective of the segmentation techniques is the partition of images or volumes in its structural components. Although its use is of basic importance in the image processing field, the effectiveness and homogeneity of the segmentation techniques still represent a problem in the area of applied engineering whose mathematical solution is yet to be found. This work focuses on the segmentation of Magnetic Resonance images and volumes of some anatomical structures related with speech production, namely, the vocal tract. Through the segmentation we were able to gather new morph-functional information on the European Portuguese production. This investigation is part of a broad project, named HERON, whose objective is the development of a speech synthesizer for European Portuguese. After the comparative analysis of an extensive set of segmentation methods, the ones that were more adequate for our application were chosen. The implemented techniques were: Region Growing; Live Wire; Level Sets; and Reaction Diffusion Active contours. The advantages and disadvantages of each method were also evaluated. Region Growing proved capable of segmenting the 2D images. For 3D information, the use of more complex methodologies, such as Level Sets, was necessary. A new "2.5D" segmentation strategy was also proposed, based on multi-planar reformatting through the longitudinal axes of the vocal tract, followed by the application of 2D segmentation algorithms as Live Wire and Level Sets. This methodology enabled the 3D Magnetic Resonance acquisitions for imaging studies in the European Portuguese production.

MSc Thesis

Segmentação do tracto vocal a partir de estudos imagiológicos de ressonância magnética 2008