Segmentation of brain mr images using wavelet transform and self organizing maps neural network

MR is a painless and noninvasive medical imaging technique in which images of bones and other tissues are taken with a special machine. It is widely used in order to obtain high quality medical images. Segmentation of brain MR images, i.e. separation of tissues of interest from other tissues, and making them clear is a very important and difficult task that is required for the detection of brain tumors and other neurological diseases. In this study, brain MR images are segmented into healthy tissues as white matter, grey matter and cerebrospinal fluid, and if any tumor and edema. In the literature, there are no studies that segments healthy brain tissues together with tumor and edema utilizing the methods used in this study.In this study, tissue segmentation of brain MR images is performed on a brain model that is generated using real brain MR images, brain MR images of healthy people and brain MR images of patients with tumor that are obtained from the hospital. Images that are filtered using anisotropic diffusion filter are aligned to the same coordinate system by registration. An algorithm that combines thresholding and morphological processes is developed to remove the areas other than the brain tissues from the images. Wavelet transform is used to obtain features from the images. Self-organizing map which is an unsupervised neural network is used for the segmentation process. Self-organizing map is labeled using the developed algorithm that is based on manually labeled image regions and hit histograms of the output neurons. Supervised learning vector quantization algorithm is used to adjust the best settlements of the neurons. Results are evaluated by sensitivity, specificity, and Dice similarity index that shows the overlap of the manual segmentation and the segmentation results obtained from the system. A graphical user interface is developed in MATLAB environment to perform the steps of the segmentation process by seeing the results of each step. Segmentation results indicate that segmentation of the brain MR images was feasible by wavelet transform and self-organizing maps neural networks.