基于多域先验的乳腺超声图像协同分割

乳腺超声(Breast ultrasound, BUS)图像具有较低的信噪比、 较低的对比度以及较模糊的边缘, 其分割是一项富有挑战性的工作. 本文提出了一种多域协同分割模型, 该模型通过结合空域与频域先验, 并引入协同分割的思想来实现对乳腺超声序列的分割. 模型在空域中得到肿瘤的姿态、 位置和强度信息, 在频域中通过使用相位一致性与零交叉检测得到肿瘤的边缘信息, 最后利用协同分割的思想构建起全局能量项, 有效地利用了图像序列信息.实验结果表明, 与传统的乳腺超声图像分割方法相比, 本文提出的分割模型能够很好地处理低对比度低回声图像以及单帧分割模型不能有效分割的图像, 分割结果具有更高的准确性.

Breast Ultrasound Image Co-segmentation by Means of Multiple-domain Knowledge

Because of low signal-noise ratio, low contrast and blurry boundaries, breast ultrasound (BUS) image segmentation is quite challenging. In this paper, a multiple-domain knowledge based co-segmentation model is proposed for BUS segmentation. It combines spatial and frequency domain prior knowledge and introduces the idea of co-segmentation to segment BUS sequence. First, tumor poses, position and intensity distribution are modeled to constrain the segmentation in the spatial domain, and then the phase feature and zero-crossing feature in the frequency domain. Finally, the BUS sequence segmentation is formulated as a co-segmentation problem. Experimental results show that the proposed method can handle low contrast and hypoechoic BUS images well and segment BUS accurately.