改进的分水岭算法在超声肿瘤图像分割中的应用

Improved Watershed Algorithm for Segmentation of Tumor Ultrasound Image

在高强度超声聚焦(HIFU)治疗中,图像自动导航是整个治疗过程的关键步骤。针对超声肿瘤图像分割提出了两种算法,分别为梯度阈值法和区域合并法。其中梯度阈值法针对分水岭过分割的缺陷,选取小于设定的梯度阈值的点作为分水岭变换的种子点,从而很好地抑制了过分割现象;区域合并法首先通过分水岭变换将图像过分割成许多具有区域均质性的超像素,然后基于最小描述(MDL)原则进行合并,将拥有相似纹理特征的小区域聚类在一起,达到抑制过分割的目的。实验结果表明这两种算法有效地解决了分水岭变换过分割的问题,同时能够很好地应用到超声肿瘤图像分割中。

In high intensity focused ultrasound（HIFU）treatment,image navigation is the key process.In this paper,two methods of gradient threshold method and region merging method were proposed for the segmentation of tumor ultrasound image.The gradient threshold method was used to select the point below the set threshold as the seed point,thus the over-segmentation phenomenon was suppressed well.The regional merging method divides the watershed transform over-divided region as super-pixel,and then uses minimum description length（MDL）principle to merge to make together the small areas with similar texture features,thus over-segmentation is suppressed.The two methods were applied to the segmentation of ultrasound tumor and good results were achieved.