Brain Tumor Detection and Classification using Geometrical Shapes as Texture Descriptors.

M.Z. Zafar; S.M. Adnan; W. Ahmad; J. Rashid; J.Ikram

No Detail Found!

Brain Tumor Detection and Classification using Geometrical Shapes as Texture Descriptors.

Author(s)

M.Z. Zafar Computer Science Department, U.E.T. Taxila, Pakistan

S.M. Adnan School of Computer Sciences, Beijing Institute of Graphics Communication, China

W. Ahmad Computer Science Department, U.E.T. Taxila, Pakistan

J. Rashid Computer Science Department, U.E.T. Taxila, Pakistan

J.Ikram School of Computer Sciences, Beijing Institute of Graphics Communication, China

Abstract

The existence of abnormal cells in the brain causes a brain tumor. There are two kinds of the tumor; Low grade or slow-growing tumor and High-grade or fast-growing tumor. Patient's successful treatment depends on the accuracy of tumor detection. Therefore, an automatic system with improved accuracy for tumor detection and classification is required. The proposed method consists of three phases to determine the presence of a brain tumor. Red, Green, and Blue (RGB) input image is converted into gray level image and skull is stripped using image masking in the preprocessing phase. In the second phase, the features are extracted using geometrical descriptors. These geometrical descriptors comprise of three geometrical shapes eclipse, parabola, and hyperbola. The performance of these geometrical shape descriptors is evaluated using Local Ternary Pattern (LTPs) and Local Quinary Patterns (LQPs). Support Vector Machine (SVM) and k-Nearest Neighbors (KNN) classifiers are used for classifying MR Images into the healthy and unhealthy brain. Experiments are performed on the Kaggle brain MR Imaging dataset and results are compared with existing techniques. Our experimental results show that the parabola descriptor achieved 97.5% as compared to other geometrical shape descriptors like eclipse and hyperbola.

Publication Details

Page(s) 83-89

DOI DOI not available

Published Journal: Technical Journal, Volume: 24, Issue: 1, Year: 2019

Request the full text from PASTIC

Keywords

Geometrical texture descriptors, Tumor classification