一种分割脑磁共振图像的改进FCM聚类算法

发布时间：2017-12-31 05:39

  本文关键词：一种分割脑磁共振图像的改进FCM聚类算法　出处：《中国生物医学工程学报》2016年06期 　论文类型：期刊论文

  更多相关文章： 脑MRI图像 噪声 偏移场 FCM算法 图像分割

【摘要】：噪声和偏移场是影响磁共振(MRI)图像质量的主要因素。以含加性噪声和乘性偏移场的脑MRI图像组织分割为目标,提出一种抗噪局部相干模糊聚类算法,通过在目标函数中加入模糊算子和一致局部信息约束,达到同时抑制噪声和偏移场不利影响的目的,提高分割准确性和稳定性。采用20例合成图像、60例来自Brain Web的模拟脑MRI图像、100例来自IBSR真实脑MRI图像,对算法的聚类性能进行评价。实验结果表明,在噪声和偏移场干扰并存的情况下,所提出算法与其他几种经典FCM改进算法相比,对合成图像集的平均分类准确度SA达到0.97,高于其他算法,最大可提高0.37;对真实脑MRI图像集的脑脊液分割有明显优势,相似性测度KI平均提高约0.1。分析表明,所提出算法有更好的分类准确性和稳定性。
[Abstract]:The noise and bias field magnetic resonance (MRI) effect is the main factors of image quality. The brain MRI images containing additive noise and multiplicative bias field segmentation as the goal, proposed an anti noise local coherent fuzzy clustering algorithm, by adding fuzzy constraint operator and consistency of bureau of the Ministry of information in the objective function, at the same time objective to suppress noise and offset the adverse effects of the field, improve the segmentation accuracy and stability. 20 cases of the use of synthetic images, simulated brain MRI images of 60 cases from Brain Web, 100 cases of IBSR from real brain MRI images, evaluate the performance of the clustering algorithm. The experimental results show that the noise and offset field interference under the condition of coexistence the proposed algorithm, and several other classical FCM algorithm compared to the average classification accuracy of the synthetic image set of SA reached 0.97, higher than that of other algorithms can be increased by 0.37; the CSF MRI image set real brain The cutting has obvious advantages. The similarity measure of KI is improved by 0.1. analysis. The results show that the proposed algorithm has better classification accuracy and stability.

【作者单位】： 大连理工大学电子信息与电气工程学部;国网辽宁省电力有限公司大连供电公司;大连医科大学附属第二医院;
【基金】：国家自然科学基金(61101230)
【分类号】：R445.2;TP391.41
【正文快照】： 引言磁共振成像(magnetic resonance imaging,MRI)是使人体内部组织结构可视化的一种成像技术,对软组织具有较清晰的成像效果。作为一种非介入检测手段,MRI已经广泛应用在临床诊疗中。脑组织的正确分割,对病灶位置、体积的确定和治疗方案的制定具有重要意义。然而,受设备性能，

本文编号：1358429