A Weighted K-means Algorithm applied to Brain Tissue Classification

• Autores: Guillermo N. Abras, Virginia Laura Ballarín
• Localización: Journal of Computer Science and Technology, ISSN-e 1666-6038, Vol. 5, Nº. 3, 2005 (Ejemplar dedicado a: Fifteenth Issue), págs. 121-126
• Idioma: inglés
• Enlaces
  • Texto completo
• Resumen

  • Tissue classification in Magnetic Resonance (MR) brain images is an important issue in the analysis of several brain dementias. This paper presents a modification of the classical K-means algorithm taking into account the number of times specific features appear in an image, employing, for that purpose, a weighted mean to calculate the centroid of every cluster. Pattern Recognition techniques allow grouping pixels based on features similarity. In this paper, multispectral gray-level intensity MR brain images are used. T1, T2 and PD-weighted images provide different and complementary information about the tissues. Segmentation is performed in order to classify each pixel of the resulting image according to four possible classes: cerebro-spinal fluid (CSF), white matter (WM), gray matter (GM) and background. T1, T2 and PD-weighted images are used as patterns. The proposed algorithm weighs the number of pixels corresponding to each set of gray levels in the feature vector. As a consequence, an automatic segmentation of the brain tissue is obtained. The algorithm provides faster results if compared with the traditional K-means, thereby retrieving complementary information from the images.