Development of an Efficient Cascade Pathological-Brain Detection System using a Median Filter and Quadratic Discriminant Analysis

Abstract

This research article proposes smart utilization of a machine learning technique to discriminate between normal and pathological brain images. The method is based on the following computational technique: a median filter is utilized for pre-processing input images; discrete wavelet transform (DWT) is utilized for feature extraction; principal component analysis (PCA) minimizes the dimensionality of the wavelet coefficients; and quadratic discriminate analysis (QDA) classifies the reduced features as normal or pathological. Experiments were carried out on 90 images (five normal and 85 pathological) from a Harvard Medical School dataset. The proposed system yielded excellent classification accuracy of 98.90% with 10× 5-fold stratified crossvalidation (SCV). Moreover, the proposed technique outperforms seven state-of-the-art algorithms in terms of accuracy. Furthermore, our method signifies its effectiveness when compared with other machine learning approaches.