Title:	TONGUE SEGMENTATION FROM MRI IMAGES USING ITK-SNAP: PRELIMINARY EVALUATION
Author(s):	Paula Martins, Catarina Oliveira, Samuel Silva, Augusto Silva, António Teixeira
ISBN:	978-972-8939-48-9
Editors:	Yingcai Xiao
Year:	2011
Edition:	Single
Keywords:	Speech Production, 3D MRI, Tongue segmentation, European Portuguese, Validation
Type:	Full Paper
First Page:	3
Last Page:	10
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	The purpose of this study was to evaluate and compare the efficiency, reliability and accuracy of manual and semi-automatic segmentation techniques to segment tongue images. This work is included in a vast framework (HERON II) that aims to improve an articulatory-based speech synthesizer (SAP-Windows), for European Portuguese (EP). Volumetric data from Magnetic resonance images (MRI) were used to extract tongue configurations from several speakers uttering different EP sounds, or the same sound produced in different contexts or syllabic positions, in a speech production study. Segmentations were performed manually and using a semi-automatic approach implemented in ITK-SNAP (Region Competition Snakes). Results from similarity metrics (Jaccard coefficient and voxelwise comparison) revealed that the semi-automatic (SA) method presents good agreement with the manual segmentation method (Jaccard=0.9002 and 10.495 voxel error). Furthermore, the semi-automatic method is more reproducible (Jaccard=0.9382 and 6.388 voxel error) than manual segmentation method (Jaccard= 0.9170 and 8.662 voxel error). The semi-automatic approach provides an efficient and reliable method to segment tongue images providing 3D visualizations that allows description and comparison of tongue configurations during the production of different sounds. This information is of great relevance in speech production field contributing to a better understanding of speech production mechanisms.
	Go Back