Integrated Approach for Brain Classification using DWT with MGLCM and autoencoder based Neural Network

Abstract

Nowadays machine learning [8] and deep learning uses in medical field increased immense level. To help to doctors and to reduce the efforts of manual testing, many authors and researchers designed algorithms which help to get outcomes automatically. Our proposed technique consists of two main algorithms; one is deep learning algorithm for feature extraction and second is handcrafted feature extraction algorithm. For deep learning algorithm it was used Convolutional neural network and for handcrafted type of feature extraction I used integration of the Modified gray level co-occurrence matrix (MGLCM) and Discrete Wavelet Transform (DWT). In handcrafted both frequency domain and spatial domain features are selected. For proposed work analysis I used run time calculation and accuracy of classification which shows the superior performance of the existing techniques as well as individual techniques. MGLCM contains modified features which include contrast, correlation, energy, entropy, homogeneity, etc. Neural network architecture is used with autoencoders. Autoencoders encodes the data in its own format for understanding the differences in normal and abnormal brain images.MRI brain scan database is used for analysis which is downloaded from free available databases.