Review of k-Nearest Neighbor advances for brain tissue segmentation

Abstract

Brain Tissue Segmentation (BTS) is used in clinical research to quantify the different types of brain tissue depicted in MR images. The segmentation can be done manually by an expert (radiologist), who points out voxels and labels them. This is very time-consuming and laborous work and is hard to reproduce. Over the years automated segmentation methods have been proposed and succesfully applied. Some of these segmentation methods use algorithms from pattern recognition. One of those methods is the k-Nearest Neighbor (k-NN ) method. The k-NN method has been succesfully applied, not only in BTS, but also other appliances. This thesis will review and compare adaptations to the k-NN method which are applicable to brain tissue segmentation. We sought methods that can improve classiffcation quality and/or improve the speed of the classiffcation. In Section 2 we will explain the k-NN method in general and, afterwards, how k-NN is applied in BTS more speciffcally. Section 3 will explain some important techniques used in the articles. In Sections 4 and 5 the advances of k-NN are reviewed and compared. In Section 6 we will propose a number of methods we think are the most promising for BTS speciffcally.