行程长度纹理特征应用于肠癌病理图片识别

Recognition of colorectal cancer pathological images based on run length texture features

传统的肠癌病理诊断都由病理医生完成,随着图像处理技术的发展,为满足医学病理图像辅助诊断的需要,提出用灰度行程纹理特征(GLRLM)来识别大肠病变切片.考虑到传统的灰度行程长度纹理特征预处理方式未充分利用图像彩色信息和病理图像的组织学信息,提出将模糊C均值应用于大肠彩色病理图像的预处理,然后提取图像的行程长度纹理特征,最后利用支持向量机分类.通过与灰度共生矩阵纹理特征对比,行程长度纹理特征和改进的行程长度纹理特征具有更高的分类准确率.同时用SVM分类器与BP神经网络、最近邻分类器对比,根据实验结果得出SVM分类器更适合小样本肠癌病理图像的分类.

Conventional pathology of colorectal cancer is diagnosed by pathologists.With the development of image processing technology,the gray level run length matrix（GLRLM）is used to recognize the pathological images in order to meet the demand of computer-aided diagnosis for medical images.Because the traditional GLRLM algorithm ignores the color and structural information behind the images,an improved algorithm was proposed using FCM for the preprocessing.Then the run length texture features of the image are extracted.Finally,the SVM is used to classify the pathological pictures.Compared with gray level co-occurrence matrix texture feature extraction algorithm,the experiments show that the traditional GLRLM and improved GLRLM have higher classification accuracy.Meanwhile,compared with KNN and BP,experiments show that SVM classifier is more appropriate for classification of small samples such as colorectal cancer pathological images.