Optimal Feature Set Selection for Brain Tumor Classification using Genetic Algorithm with Support Vector Machine and Decision Tree

In the recent years, disease identification and classification from medical imaging plays an important role in timely diagnosis and treatment. This paper focuses on classifying the axial Magnetic Resonance Images (MRI) of brain tumor into four types, namely Glioma, Meningioma, Metastasis and Normal. The proposed approach integrates the Statistical and Textural Feature Extraction (STFE) methods, Genetic optimization Algorithm (GA) with multiclass classifiers-Support Vector Machine (SVM) or Decision Tree (DT). STFE methods are applied to the training images and 26 features are extracted. From the extracted features, irrelevant features are removed using GA and optimal feature set is identified. The optimal feature set is applied to multiclass SVM/DT for brain tumor type classification. Four different experiments namely SVM, SVM with GA (SVMGA), DT and DT with GA (DTGA) are executed and their results are analyzed. It has been inferred that the performance of SVMGA and the DTGA are better than SVM and DT with an increased accuracy of 20.69% and 15% respectively. DTGA method resulted with a maximum accuracy of 95% among the four different approaches.