A Brain MR Images Segmentation Method Based on SOM Neural Network

• Published in: 2007 1st International Conference on Bioinformatics and Biomedical Engineering Vol. 1; pp. 686 - 689
• Main Authors: Tian, D., Fan, L.
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 2007
• Subjects: Biological neural networks; Clinical diagnosis; Feature extraction; Humans; Image segmentation; Magnetic resonance; Nearest neighbor searches; Noise level; Noise robustness; Visualization
• ISBN: 9781424411207; 1424411203
• ISSN: 2151-7614
• DOI: 10.1109/ICBBE.2007.179

Image segmentation is an indispensable process in the visualization of human tissues, particularly during clinical analysis of magnetic resonance (MR) images. In this paper, a novel brain MR images segmentation method is presented based on self-organizing map (SOM) neural network. The method comprises two main steps: feature extraction and pixel classification based on SOM neural network. In traditional techniques, neural network's input is the feature vector extracted from the intensity of the pixel and of its n nearest neighbors, which introduces dependency on the gray levels spatial distribution, and thus the final segmentation results are prone to be effected by noise. To enhance the robustness of the method, we perform statistical transformation to the traditional feature vector as neural network's input. Simulated brain MR images with different noise levels and intensity inhomogeneities are segmented to demonstrate the superiority of the proposed method compared to the traditional technique.