结合灰度共生特征和水平集算法的宫颈醋白区域分割

Acetowhite Region Segmentation in Cervix Based on Gray Level Co-occurrence Characteristic and Level Set Algorithm

肉眼醋酸实验是宫颈癌筛查的重要手段,使阴道镜设备具备自动识别醋白区域的功能是解决在临床上缺乏有经验医生这一难题的有效方法。针对这一目的,提出了一种建立在灰度共生特征矩基础上的CV模型水平集算法。该方法首先使用k-means聚类从肉眼醋酸实验后的原始宫颈图像中分割出宫颈区域,继而利用合成的灰度共生特征矩对宫颈区域进行醋白特征提取并获得待分割的特征图,最后使用改进的CV水平集算法对特征图进行分割并得到醋白区域。实验结果显示:改进后的CV水平集算法比传统CV水平集算法的敏感度在平均值上低26.6%,比分水岭分割高47.6%,比模糊聚类分割高11.23%;其特异性在平均值上比水平集分割高29.45%,比分水岭分割低11.64%,比模糊聚类高45.23%;而以Jaccard Index(JI)统计的精度指标在平均值上比传统CV水平集算法高19.74%,比分水岭算法高23.27%,比模糊聚类高38.11%。该新方法在总体性能指标上精度更高。

Naked eye acetic acid test is an important means of cervical cancer screening,so that the automatic identification of the white area in the colposcopy equipment is an effective way to solve the problem of lack of experienced doctors in clinic. Aiming at this purpose,an improved CV model level set algorithm based on gray level co-occurrence characteristic matrix was proposed in this paper.Firstly,the cervical region was segmented by using k-means algorithm from the original post-acetic acid test cervix image. Secondly,a composite gray level co-occurrence moment characteristic was used to extract the acetowhite（ AW） feature and configure the feature image to be segmented. Lastly,a modified CV model level set algorithm was used to segment the feature image and the AW region was obtained eventually. The experimental results show that the modified level set algorithm gains an average 26. 6% lower sensitivity and an average 29. 45% higher specificity comparing with the original CV model level set algorithm. It also gains an average 47. 6%higher sensitivity and an average 11. 64% lower specificity comparing with watershed algorithm and an average 11. 23% higher sensitivity and an average 45. 23% higher specificity comparing with fuzzy clustering algorithm. However,the developed method has an average 19. 74%,an average 23. 27% and an average 38. 11% higher JI（ Jaccard Index） accuracy separately comparing with the three aforementioned algorithms. It can be concluded that the new method is a more accurate algorithm in the overall performance.