Title: A Novel Coarse-to-fine Level Set Framework for Ultrasound Image Segmentation

Year of Publication: 2015
Page Numbers: 8-17
Authors: Jinze Yu, Pheng-Ann Heng, Weiming Wang, Jing Qin
Conference Name: The Second International Conference on Artificial Intelligence and Pattern Recognition (AIPR2015)
- China


Ultrasound image segmentation is a fundamental but undoubtedly challenging problem in many medical applications due to various unpleasant artifacts, e.g., noise, low contrast and intensity inhomogeneity. This paper presents a coarse-to-fine framework for ultrasound image segmentation based on a preprocessing step via speckle reducing anisotropic diffusion (SRAD) and a modified version of Chan-Vese model by proposing novel evolution functional involving the Sobolev gradient. SRAD is a diffusion method tailored for ultrasound image denoising, and is adopted here to construct a despeckled image which allows us to obtain a coarse segmentation of the input image by carrying out our proposed CV model. This coarse segmentation will be further used by our level set model as a constraint to guide the fine segmentation. We compare the proposed model with some famous region-based level set methods. Experimental results in both synthetic and clinical ultrasound images validate the high accuracy and robustness of our approach, indicating its potential for practical applications in ultrasound imaging.