Selection of initial parameters of K-means clustering algorithm for MRI brain image segmentation

To solve the problem of classification number and how to select the initial clustering center to segment magnetic resonance imaging (MRI) brain image by using K-means clustering algorithm, this paper proposes a new strategy to get initial clustering center of white matter (WM), gray matter (GM) and cerebrospinal fluid (CSF), background (BG) by using moving average filtering method or gray matrix normalization method. This paper also discusses problem of classification number by analyzing their clustering centers and combining clustering centers from the perspective of qualitative and quantitative. The experimental results show that MRI brain image divided into 4 classes is reasonable and selection of initial cluster centers by using gray matrix normalization method for brain tissue segmentation is effective, which effectively improve the computer efficiency compared with the traditional K-means algorithm, saving more than 30% of the running time.