乳腺超声图像中的肿瘤区域定位研究

Study on the regional localization of tumor in breast ultrasound image

采用先验知识模型定位乳腺超声图像中的肿瘤区域时,对脂肪组织、伪影等干扰因素的抵抗力差,当肿瘤位置发生变化时无法自动定位肿瘤区域,定位肿瘤区域准确率低。提出新的乳腺超声图像中的肿瘤区域定位方法,利用形态学掩模对乳腺超声图像进行滤波,采用各向异性扩散方法降低图像中的噪声,获取最佳乳腺超声图像;提取最佳乳腺超声图像灰度共生矩阵中的纹理特征;通过全自动肿瘤定位方法,采用基于图像灰度共生矩阵纹理特征的自动参考点选择方法得到自动参考点,在此基础上通过Mean-Shift算法迭代获取种子点,该点则是肿瘤所处区域。实验结果表明,所提出方法定位恶性乳腺肿瘤准确率为100%,定位良性乳腺肿瘤准确率高于97%,具有较高的定位精度。

When using a prior knowledge model to locate the tumor region in the ultrasound image of the breast, the resistance to interference factors such as adipose tissue and pseudoimage is poor. When the tumor position chan ges ,it is impossible to automatically locate the tumor region, and the accuracy of locating the tumor region is low. In this paper, a new method of tumor region location in breast ultrasound images is proposed. The morphological mask is used to filter breast ultrasound images. The anisotropic diffusion method is used to reduce the noise in the image and obtain the best breast ultrasound images. And extract the texture characteristics of the optimal breast ultrasound image gray scale symbiosis matrix. Automatic reference point based on gray-scale symbiosis matrix texture selection method is adopted by automatic tumor location method. Based on this, the seed point is obtained by Mean-Shift algorithm iter atively. This point is the area where the tumor is located. The experimental results show that the method has 100% ac curacy in locating malignant mammary tumors and 97% accuracy in locating benign mammary tumors.