A clustering fusion technique for MR brain tissue segmentation

• Published in: Neurocomputing (Amsterdam) 2018-01, Vol.275, p.546-559
• Main Authors: Al-Dmour, Hayat, Al-Ani, Ahmed
• Format: Article
• Language: English
• Subjects: Fuzzy c-means; k-means; Neural network; Segmentation; self-organized map; Superpixel
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2017.08.051

•An automatic method for segmenting brain image into three regions is proposed.•The method is based on superpixel, three clustering techniques, and neural network.•Neural Network is used to imitate the clustering and reduce the computational cost.•The segmented image is combined using multiple clustering algorithms.•Our method outperformed the three base clustering techniques. In recent decades, a large number of segmentation methods have been introduced and applied to magnetic resonance (MR) brain image analysis to measure and visualize the anatomical structures of interest. In this paper, an efficient fully-automatic brain tissue segmentation algorithm based on a clustering fusion technique is presented. In the training phase of this algorithm, the pixel intensity value is scaled to enhance the contrast of the image. The brain image pixels that have similar intensity are then grouped into objects using a superpixel algorithm. Further, three clustering techniques are utilized to segment each object. For each clustering technique, a neural network (NN) model is fed with features extracted from the image objects and is trained using the labels produced by that clustering technique. In the testing phase, pre-processing step includes scaling and resizing the brain image are applied then the superpixel algorithm partitions the image into multiple objects (similar to the training phase). The three trained neural network models are then used to predict the respective class of each object and the obtained classes are combined using majority voting. The efficiency of the proposed method is demonstrated on various MR brain images and compared with the three base clustering techniques.