Multi-level 3D Wavelet Analysis: Application to Brain Tumor Classification

Abstract

A variety of approaches to achieve automatic grading of brain tumors has surfaced in the recent past. In this paper, we propose a technique for automatic classification of brain tumor based on multi-level 3D wavelet analysis. A wavelet analysis is capable of performing multiresolution analysis under different environments i.e. it is rotation and direction invariant. The brain tumor MR images are decomposed using wavelet transform and the approximation and detail coefficients at each level are used as feature vectors after dimensionality reduction. The performance of the system is evaluated using four state-of-the-art classifiers namely Support Vector Machine, Multi-layer Perceptron, Meta Multi Class and Random Forest. All experiments are performed on BRATS 2015, a benchmark database for brain tumor MR images. We have achieved promising results with highest accuracy of 99.3% for sym4 wavelet function.