Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (3): 187-193.DOI: 10.3778/j.issn.1002-8331.1810-0220

Previous Articles     Next Articles

Segmentation and Feature Extraction of Brain Tumor Based on Magnetic Resonance Image Using K-means

ZONG Xiaoping, TIAN Weiqian   

  1. College of Electronic Information Engineering, Hebei University, Baoding, Hebei 071000, China
  • Online:2020-02-01 Published:2020-01-20

采用K-means的脑肿瘤磁共振图像分割与特征提取

宗晓萍,田伟倩   

  1. 河北大学 电子信息工程学院,河北 保定 071000

Abstract: Brain tumor segmentation is important for physicians to determine the extent of tumor progression. However, due to the irregular shape of the tumor, low contrast with surrounding tissues, and the location of the tumor is not fixed, the precise segmentation of brain tumors is very difficult. The traditional [k]-means segmentation method only utilizes the image’s grayscale features, so it is difficult to accurately segment the tumor boundary. In this paper, texture features produced by gray symbiosis matrix are used and [k]-means algorithm features are combined with geometric invariant moments to segment brain MRI images. The gray co-occurrence matrix, which is defined as the joint probability distribution of pixel pairs, is a symmetric matrix. It not only reflects the comprehensive information of the image’s gray level in the adjacent directions, adjacent intervals and variation range, but also reflects the location distribution features between the same gray level pixels, which is the basis for calculating texture features. Geometric moment(invariant moment) is characterized by rotation, translation, scale, etc., which can decompose the image into finite feature values. Moreover, by comparing the invariant moment parameters of the tumor image extracted from the same patient, the degree of geometric shape change of the tumor can be obtained. Experimental results show that this method can effectively improve the segmentation accuracy of brain tumor fringe region.

Key words: K-means, feature extraction, gray level co-occurrence matrix, invariant moment, correlation coefficient

摘要: 大脑肿瘤分割对于医师判断肿瘤恶化程度非常重要。然而,由于肿瘤的不规则形状、与周围组织的低对比度以及出现位置的不固定,给脑肿瘤的精确分割带来很大的困难。传统的K-means分割方法仅仅利用图像的灰度特征,很难准确分割肿瘤边界。利用灰度共生矩阵提取出的纹理特征,并结合图像几何不变矩特征对分割出的脑肿瘤图像进行特征提取。灰度共生矩阵定义为像素对的联合概率分布,是一个对称矩阵,它不仅反映图像灰度在相邻的方向、相邻间隔、变化幅度的综合信息,也反映了相同的灰度级像素之间的位置分布特征,是计算纹理特征的基础;几何矩(不变矩)具有旋转、平移、尺度等特性,能将图像分解为有限特征值,并且通过对比所提取出的同一病人的肿瘤图像的不变矩参数,可以获得该肿瘤几何形状变化程度。实验结果表明,该方法可以同时从纹理和几何特征对图像特征进行描述,与分别采用灰度共生矩阵和不变矩方法进行特征提取相比较,降低了算法计算量,同时提升了算法的抗噪性。

关键词: K-means, 特征提取, 灰度共生矩阵, 不变矩, 相关系数