Abstract: In the process of breast cancer radiotherapy, the accuracy of mammary gland segmentation is directly related to the effectiveness of patient’s treatment. A coarse-to-fine strategy is proposed for breast gland segmentation. Firstly, morphological reconstruction filter is used to remove image’s small burrs and irregular edges so as to avert new boundaries and boundary migration. The mammary gland is then marked by the area size to differentiate the foreground from the background. After labeling, minima imposition is applied to modify the gradient image. Finally, standard watershed transform is employed to implement segmentation on the basis of modified gradient image, realizing the coarse segmentation of mammary gland. For the purpose of accomplishing the fine segmentation, the curve formed by selected seed points on the watershed ridge is chosen as the initial contour of the snake model to dynamically approach target’s real edges. Experiments show that this method has higher accuracy for mammary gland segmentation.

Key words: mammary gland, morphological reconstruction filers, marker-controlled watershed, Snake model

摘要： 在乳腺癌放射治疗过程中，乳腺腺体分割的准确与否直接关系到患者的治疗效果，提出了一种从粗到细的乳腺腺体图像分割方法。首先利用形态学重构滤波器去除图像中的细小毛刺和不规则边缘，避免产生新边界和边界偏移；然后通过面积标记乳腺腺体，采用强制最小技术修正梯度图像并进行分水岭变换，实现乳腺腺体的粗分割；通过在分水岭脊线上人工选择种子点形成的曲线作为Snake模型的初始轮廓来动态逼近目标的真实边缘，实现乳腺腺体的细分割。实验结果表明，该方法对乳腺腺体的分割具有较高的准确性。

关键词: 乳腺腺体, 形态学重构滤波器, 标记控制分水岭, Snake模型