MRI Brain Abnormalities Segmentation using K-Nearest Neighbors (k-NN)

Segmentation of medical imagery remains as a challenging task due to complexity of medical images. This study proposes a method of k-Nearest Neighbor (k-NN) in abnormalities segmentation of Magnetic Resonance Imaging (MRI) brain images. A preliminary data analysis is performed to analyze the characteristics for each brain component of “membrane”, “ventricles”, “light abnormality” and “dark abnormality” by extracting the minimum, maximum and mean grey level pixel values. The segmentation is done by executing five steps of k-NN which are determination of k value, calculation of Euclidian distances objective function, sortation of minimum distance, assignment of majority class, and determination of class based on majority ranking. The k-NN segmentation performances is tested to hundred and fifty controlled testing data which designed by cutting various shapes and size of various abnormalities and pasting it onto normal brain tissues. The tissues are divided into three categories of “low”, “medium” and “high” based on the grey level pixel value intensities. The overall experimental result returns good and promising segmentation outcomes for both light and dark abnormalities. Keywords-k-Nearest Neighbor (k-NN); brain abnormalities segmentation, Magnetic Resonance Imaging (MRI)