摘要
目的提出一种基于偏场校正的概率模糊C均值(BCPFCM)改进算法,并应用于脑部MRI图像分割。方法采用BrainWeb数据库中2幅人工合成图像和20幅脑部MRI图像进行仿真实验。第一步,构建MRI图像信息模型,并分离出偏场图像。第二步,采用像素加权隶属度和加权一致性测度解决难以分类的邻域空间异质性像素。第三步,构建基于模糊C均值算法、BCPFCM改进算法目标函数。采用误分类率和Tanimoto指数,比较不同算法图像分割结果。结果BCPFCM改进算法能准确检测高低频条带与像素灰度不均等偏场信息,并完整分割出人工合成图像与脑部MRI图像各功能分区。基于BCPFCM改进算法的人工合成图像分割精度为100%;含40%灰度不均偏场的脑部MRI图像误分类率为5.74%,含1%~5%水平噪声图像误分类率为6.85%~14.34%。基于BCPFCM改进算法分割所得白质、灰质和脑脊液Tanimoto指数分别为0.75±0.03、0.79±0.04和0.39±0.01。结论 BCPFCM改进算法分割精度高、鲁棒性强,是一种可行的脑部MRI图像分割算法。
Objective To propose modified fuzzy C-means algorithm based on bias field corrected possibility fuzzy C-means(BCPFCM) and apply to brain magnetic resonance imaging(MRI) image segmentation. Methods Two artificially synthesized images and 20 brain MRI images in BrainWeb database were used for simulation experiments. First, the MRI image information model was constructed estimating embedded bias field. Second, the difficult-to-classify neighborhood spatial heterogeneity pixels were resolved by pixel weighted membership and weighted consistency measures. Third, the objective function was constructed based on fuzzy C-means algorithm and modified BCPFCM algorithm. The image segmentation results of different algorithms were compared by misclassification rate and Tanimoto index. Results The BCPFCM algorithm was successfully estimated bias field of images with low-frequency, high-frequency and spatial inhomogeneity, the synthetic images and different tissue classes of MRI images were accurately segmented. The segmentation accuracy of artificially synthesized images based on modified BCPFCM algorithm was 100 %. The misclassification rate of brain MRI images with 40 % gray uneven bias field was5.74 %, and images with 1 %-5 % level noise were 6.85 %-14.34 %. The Tanimoto index of white matter, gray matter and cerebrospinal fluid by modified BCPFCM algorithm was 0.75 ± 0.03, 0.79 ± 0.04 and 0.39 ± 0.01, respectively. Conclusion It is demonstrated that modified BCPFCM algorithm showed high segmentation precision and strong robustness, which is a feasible brain MRI image segmentation algorithm.
作者
黄明英
唐文伟
张涛
何昌荣
郭兰
HUANG Ming-ying;TANG Wen-wei;ZHANG Tao;HE Chang-rong;GUO Lan(Department of Radiology,Women’s Hospital of Nanjing Medical University,Nanjing Maternity and Child Health Care Hospital,Nanjing 210004,Jiangsu,China)
出处
《生物医学工程与临床》
CAS
2021年第3期277-283,共7页
Biomedical Engineering and Clinical Medicine
关键词
图像分割
MRI
模糊C均值
偏场校正
改进算法
邻域
image segmentation
MRI
fuzzy C-means
bias field correction
modified algorithm
neighborhood
