| 摘要: |
| 传统的模糊C-均值聚类（FCM）算法只考虑了图像灰度信息，未考虑图像的邻域信息，抗噪性能不够理想.为了充分利用图像空间信息，提出一种结合马尔可夫随机场（MRF）的自适应加权FCM改进算法.该算法根据局部密度判断像素在其窗口邻域范围内的离散种类，将MRF空间约束场和隶属度场的权重根据像素离散种类进行自适应变化，在消除噪声影响的同时，尽可能保留弥散张量成像（DTI）的图像细节信息.实验结果表明：该算法可以准确分割DTI图像，得到边缘清晰且细节信息保留良好的分割结果，与FCM算法以及MRF和FCM融合算法相比，其分割系数至少提高了3%，分割熵至少降低了2%，分割聚类效果得到提高，且分割系数和分割熵都不易受噪声幅度的影响. |
| 关键词: 模糊C-均值聚类(FCM) 医学图像分割 马尔可夫随机场(MRF) 弥散张量成像（DTI）图像 离群点检测 自适应权重 |
| DOI：10.3969/J.ISSN.1000-5137.2020.01.008 |
| 分类号:TP391.4 |
| 基金项目:国家自然科学基金（61862029） |
|
| DTI image segmentation algorithm based on Markov random field and fuzzy C-means clustering |
|
CHEN Kang, ZHANG Xiangfen, MA Yan, YUAN Feiniu, LI Chuanjiang
|
|
College of Information, Mechanical and Electrical Engineering, Shanghai Normal University, Shanghai 201418, China
|
| Abstract: |
| The traditional fuzzy C-means clustering (FCM) algorithm only considered the gray information of the image,not including the neighborhood information of it,which would lead to an unsatisfactory anti-noise performance. In order to make full use of image space information,an improved adaptive weighted FCM algorithm combining with Markov random fields (MRF) was proposed in this paper. According to the local density,the discrete types of pixels in the neighborhood of the window were estimated,and the weights of MRF spatial constraint field and membership field were changed adaptively according to the discrete types of pixels,so as to eliminate the influence of noise and maintain the diffusion tensor imaging（DTI）image details as much as possible. The experimental results showed that this algorithm could segment DTI image accurately and achieve the segmentation with clear edge and satisfying detail information reservation. Compared with FCM algorithm and existing MRF and FCM fusion algorithm,the segmentation coefficient was improved by at least 3%,the segmentation entropy was reduced by at least 2%. At the same the segmentation clustering effect was improved,and the segmentation coefficient and the segmentation entropy were not easily affected by the noise amplitude. |
| Key words: fuzzy C-means clustering (FCM) medical image segmentation Markov random field (MRF) diffusion tensor imaging（DTI）image outlier detection adaptive weight |
