Selection and Ranking of Brain Tumor MRI Image Features

Authors

  • Ameen Mohammed Abd-Alsalam Kirkuk University

Abstract

This paper presents a study about the features that are taken from abnormal (cancerous) brain MRI images, and how these features differ in value between the normal and the abnormal segments of the brain MRI images. The proposed algorithm depends on gathering abnormal brain MRI (axial view) images, applying some preprocessing as: Removing unnecessary portions, Normalizing sizes, Enhancing images through filters then dividing the image into small normal and abnormal sub segments with a size of 32 by 32 pixels ready to be feature extracted. In this paper 15 image features are extracted and studied to decide and classify the normal and abnormal sub segments and also to select the most effecting features and finally to rank them. The types of the features are: Histogram, Transform, Texture, and Statistical features. These features are leveled to 3 levels: Low, Medium, and High as (L, M, H) according to their values and are calculated in normal and abnormal sub segments to make a statistical study about their existence in both normal and abnormal sub segments. A total of 69 abnormal MRI images were taken from 11 ill persons, containing 444 normal and 345 abnormal sub segments, making a total of 789 sub segments to be feature extracted and studied. The paper shows that the features of (mean, variance, standard deviation, entropy, vertical, energy, homogeneity, kurtosis, mode and RMS roughness) were affected by the existence of tumors, while the features (horizontal, diagonal, contrast, correlation, skewness) had no effect.

Downloads

Published

2018-03-28

Issue

Section

Articles