A Novel Clustering Based Segmentation of Multispectral Magnetic Resonance Images

The application of image processing techniques has rapidly increased in recent years. Medical images almost are stored and represented digitally [26]. Medical image segmentation has very important rule in many computer aided diagnostic tools. These tools could save clinicians time by simplifying the time consuming process [27]. The brain images segmentation is a complicated and challenging task. However, accurate segmentation of these images is very important for detecting tumors, edema, and necrotic tissues. Moreover, accurate detecting of these tissues is very important in diagnosis systems. Data acquisition, processing and visualization techniques facilitate diagnosis. Image segmentation is an established necessity for an improved analysis of Magnetic Resonance (MR) images. Segmentation from MR images may aid in tumor treatment by tracking the progress of tumor growth and shrinkage. The advantages of Magnetic Resonance Imaging are that the spatial resolution is high and provides detailed images. Functional Magnetic Resonance Imaging data are a major challenge to any image processing software because of the huge amount of image voxels [8]. Magnetic Resonance Imaging has proved to provide high quality medical images and is widely used especially for brain [9]. The various MR image slices of the brain are recorded depending on the tasks the patient is performing. The MR feature images used for the segmentation consist of three weighted images namely T1, T2 and Proton Density (PD) for each axial slice through the head. In this paper, a novel algorithm is presented for unsupervised segmentation of multi-spectral images, based on the research, through neural network techniques, of an optimized space in which to perform clustering. Tests performed on both real and simulated MR images shows good result, encouraging the application to different medical targets and further investigation.