A Fast K-means Clustering Algorithm for Separation of Brain Tissues in MRI

Efficient separation of brain tissues in Magnetic Resonance Imaging (MRI) is a very important step for the quantitative diagnosis of brain diseases. To identify important brain regions such as the gray matter (GM), white matter (WM) and the cerebrospinal substance fluid spaces (CSF), we proposed in this paper a new improved K-means algorithm (called HKM: Histogram-based K-Means) based on the image histogram and the median filter. The proposed algorithm is characterized by its ability to segment image faster and robustness in the presence of noise and non-uniform tissues. Moreover, it reduces the computational complexity and improves the performances of K-means algorithm in terms of the Jaccard and Dice Indexes. The obtained results demonstrate that the proposed HKM algorithm requires less times and achieves better results than the standard K-means clustering algorithm.