多特征聚类与粘连分离模型的细胞抹片图像分割与分类

发布时间：2018-05-17 10:49

本文选题：胰腺细胞 + 细胞抹片显微图像分割　；参考：《生物医学工程学杂志》2017年04期

【摘要】：胰腺癌的诊断非常重要,而细胞抹片显微图像的病理分析是其诊断的主要手段。图像的准确自动分割和分类是病理分析的重要环节,因此本文提出了一种新的胰腺细胞抹片显微图像自动分割与分类算法。在分割方面,首先采用多特征Mean-shift聚类算法(MFMS)定位细胞核区域;接着采用弹性数学形态学结合角点检测的去粘连模型(CSM)对粘连重叠细胞核进行去粘连处理,实现了分割的准确性和鲁棒性。在分类方面,首先针对分割的细胞核提取了4个形状特征和138个不同颜色空间的纹理特征;然后结合支持向量机(SVM)和链式遗传算法(CAGA)实现封装式特征选择;最后将优选特征送入SVM进行分类,完成了胰腺细胞抹片显微图像的分类识别。本文采用了15幅图像一共461个细胞核进行测试。实验结果显示,本文算法可以实现不同类型的胰腺细胞抹片显微图像的自动分割与准确分类。就分割来说,本文算法可获得较高的正确率(93.46%±7.24%);就正常和癌变细胞的分类来说,本文算法可获得较高的分类正确率(96.55%±0.99%)、灵敏度(96.10%±3.08%)和特异度(96.80%±1.48%)。
[Abstract]:The diagnosis of pancreatic cancer is very important, and the pathological analysis of cell smear microscopic image is the main means of diagnosis. Accurate automatic segmentation and classification of images is an important part of pathological analysis. Therefore, a new automatic segmentation and classification algorithm for pancreatic cell smear microimages is proposed in this paper. In the aspect of segmentation, firstly, multi-feature Mean-shift clustering algorithm was used to locate the nuclear region, and then the adhesion removal model was used to deal with the superimposed nuclei by elastic mathematical morphology combined with corner detection. The accuracy and robustness of segmentation are realized. In classification, four shape features and 138 texture features in different color spaces are extracted from the segmented nuclei, and then the encapsulated feature selection is realized by combining support vector machine (SVM) and chain genetic algorithm (CAGA). Finally, the optimal selection features were sent into SVM for classification, and the classification and recognition of pancreatic cell smear microimages were completed. A total of 461 nuclei were tested in 15 images. The experimental results show that the proposed algorithm can realize automatic segmentation and accurate classification of different types of pancreatic cell smear microimages. As far as segmentation is concerned, the proposed algorithm can obtain a higher accuracy rate of 93.46% 卤7.24% and a higher accuracy rate of 96.55% 卤0.99% and a sensitivity of 96.10% 卤3.08% and a specificity of 96.80% 卤1.48% for the classification of normal and cancerous cells.
【作者单位】：重庆大学通信工程学院;第三军医大学生物医学工程学院;
【分类号】：R735.9;TP391.41

【相似文献】