Tongue Segmentation From MRI With Geometrically Constrained Snake Model

In the field of speech generation mechanism research,the non-invasive and nonhazardous aspects of magnetic resonance imaging(MRI)provide a possibility for pronunciation visualization.The purpose of pronunciation visualization is to accompany the generation of speech signals,to provide accurate vision for speech organs(tongue,lips,and velum),to help people understand the pronunciation mechanism and correct pronunciation,and also to help the rehabilitation of patients with speech disorders and pronunciation correction.Therefore,pronunciation visualization is crucial in basic research and some specific applications.To construct a pronunciation visualization system,the outline of speech organs must be extracted from images collected by various medical devices for further three-dimensional modeling.As a flexible speech organ,the tongue deforms and moves during the process of speech production,often touching other tissues in the surrounding area.In addition,the quality of MRI images is usually not high,which makes the task of accurately segmenting the tongue difficult.The traditional image segmentation algorithms are mostly based on the characteristics of the image itself,which cannot effectively solve this problem.This paper uses three kinds of MRI databases,covering English,Japanese and Chinese.First,several typical algorithms are used for comparison.Then the anatomical information of the tongue is used as the high-latitude feature for geometric constraints,and the Snake model is improved.The tongue contour segmentation for the mid-sagittal MRI is used to deal with the unevenness of the tongue in contact with the surrounding tissues and its own distribution,which greatly improves the accuracy of tongue segmentation.The paper provides a complete visual display,and qualitative and quantitative analysis of the experiment,which proves the feasibility of the algorithm.Compared with the traditional Snake model,the segmentation performance is significantly improved,which provides a new direction for tongue segmentation.