基于曲线波的超声图像分割

Ultrasound image segmentation based on curvelet

为了提高前列腺超声图像分割的准确率,提出一种基于曲线波的半监督超声图像自动分割方法.首先,采用对微小波动敏感度高的Riemann-Liouville(RL)分数阶微分算子,突出模糊边界并增强超声图像的纹理;其次,运用曲线波变换对超声图像进行频域中的分解,获得不同子带分量以表达超声图像特征;然后,基于Adaboost的分类算法识别出超声图像中的病灶区和非病灶区;最后,采用中值滤波和腐蚀的方法使病灶区域边缘完整、平滑.实验表明,与运用共生矩阵及二进小波作纹理分析的分割结果比较,所提出的方法在准确率上有了很大的改进,分割超声图像效果更佳.

In order to improve the accuracy of prostate ultrasound image segmentation, a semi-super- vised automatic segmentation method based on curvelet transform is proposed. First, the Riemann- Liouville （RL） fractional differential operator which is sensitive to the tiny fluctuations is used to en- hance the fuzzy boundary and image texture. Secondly, the image is transformed into curvelet do- main and different subbands are obtained to represent the ultrasound image characteristics. Thirdly, the Adaboost algorithm is applied to identify the lesion and non-lesion regions in the ultrasound im- age. Finally, the median filter and the erosion operator are used to smooth the lesion regions＇ edge. Experiments show that the proposed method outperforms the approaches based on co-occurrence ma- trix and dyadic wavelet in terms of accuracy.