超声图像多囊卵巢分割及其在自动检测中的应用

Polycystic Ovary Segmentation in Ultrasound Images and its Automatic Recognition Application

医生通过人工计数卵巢超声图像上小囊胞的个数来诊断多囊卵巢综合症,存在易受人为因素干扰、重复性差、效率低等问题,为此提出一种基于超声图像的自动检测多囊卵巢综合症的方案。先以自适应形态学滤波去除卵巢超声图像的斑点噪声,接着采用改进的带标记分水岭算法提取目标(含小囊胞和类似小囊胞)轮廓,最后通过聚类方法识别出卵巢内的小囊胞。实验以专家的人工结果为标准对方案进行验证,同时与以boundary vector Field(BVF)活动模型提取卵巢轮廓进行识别的方法进行比较。结果表明,方案对多囊卵巢内小囊胞的识别正确率达到84%,比BVF方法高23%,因此能一定程度用于超声图像多囊卵巢的自动识别。

The traditional diagnosis of polycystic ovary syndrome （PCOS） performed by doctors is to manually count the number of follicular cysts in the ovary, which may lead to variability, poor reproducibility and low efficiency. In order to overcome these problems, an automatic scheme was proposed. Firstly an adaptive morphological filter was used to despeckle the ultrasound images of polycystic ovary. Then a modified labeled watershed algorithm was applied to extract contours of targets （ follicular cysts and their similar objects）. Finally a clustering method was used to identify the expected follicular cysts. Based on the standard of experts＇ results, the proposed scheme was verified, and its performance was also compared with the recognition method using the boundary vector field （BVF） active contour to extract the ovary contour. It was showed that the proposed scheme achieved the accuracy rate of 84%, which was 23 % higher than that of BVF method. Therefore, it may effectively complete the automatic recognition of the PCOS.