一种分割脑磁共振图像的改进FCM聚类算法

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BrainWeb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。

A Modified FCM Clustering Method for Brain Magnetic Resonance Image Segmentation

The noise and bias field are main factors lowering the quality of the magnetic resonance imaging. In order to segment brain tissue from MRI image, an anti-noise coherent local intensity fuzzy clustering algorithm (ANCLIFC) was proposed in this wok. By adding a new fuzzy operator and coherent local information as constraints in the cost function, ANCLIFC algorithm exhibited good clustering performance in resisting noise and bias field simultaneously. Twenty synthetic images, 20 simulated brain MRI images from BrainWeb and 100 real brain MRI images from IBSR database were used to evaluate the algorithm′s clustering performance. The experimental results demonstrated that ANCLIFC algorithm had better classification accuracy and stability than other classical modified FCM algorithms for low quality images contaminated by noise and bias field. For synthetic images, the average overall classification accuracy′s SA was 0.97, larger than other algorithms and the best improvement achieved 0.37. For real brain MRI images, ANCLIFC algorithm exhibits obvious superiority in segmenting CSF and the similarity measure′s KI increases about 0.1 in average.