基于Zernike矩和水平集的超声图像分割
Ultrasound image segmentation based on Zernike moment and level set

为了提高超声图像的分割精确率,提出了一种基于Zernike矩和水平集的超声图像分割方法. 首先,利用9个具有不同阶数和重复度的Zernike矩提取超声图像的纹理特征,保留矩的幅值和相位,获得18个特征图,同时在每一特征图目标区域内外采样,利用采样值计算出特征图的权值.然后,将特征图与高斯算子进行卷积,计算其边缘检测函数,将所有特征图的边缘检测函数与对应的特征图权值相乘,所得结果之和即为该超声图像的边缘检测函数.最后,利用基于变分函数的水平集方法对超声图像进行分割. 基于前列腺超声图像的实验结果显示,相比基于梯度的水平集方法和基于Zernike矩幅值的水平集方法,所提方法具有更高的分割精度,dice相似系数达到95%以上.
To improve the segmentation accuracy of ultrasonic images, an ultrasonic image segmentation method based on the Zernike moments(ZMs)and the level set is presented. First, 9 ZMs with different orders and repetitions are used to extract the image features. Both the magnitudes and phases are reserved to obtain 18 feature images. Meanwhile, the weights of the feature images are calculated according to the samples obtained by sampling inside and outside of the target region of each feature image. Then, the edge indicator functions are calculated by the convolution of the feature images and the Gaussian operator. The sum of the multiplication results of the edge indicator functions and the corresponding weights of the feature images is the edge indicator function of the ultrasonic image. Finally, the ultrasonic image is segmented by the level set method based on the variation formulation. The experimental results of prostate ultrasonic images show that compared with the level set method based on the gradient and the level set method based on the ZM magnitude, the proposed method has higher segmentation accuracy, and the dice similarity coefficients are more than 95%