A Bio Inspired Hybrid Krill Herd-Extreme Learning Machine Network Based on LBP and GLCM for Brain Cancer Tissue Taxonomy

Abstract

Brain cancers are the second most common disease in children. The radiologist plays a vital role in diagnosing a disease. Manual classification is a time consuming process and can cause human errors. Our objective is to develop a fully automated classification method for identification of brain cancers. Methods: This paper proposes a Bio Inspired Hybrid Krill Herd-Extreme Learning Machine (ELM) Network which classifies the Brain images into one of the classes namely normal image, Astrocytoma cancer, Meningioma cancer or Oligidendroglioma cancer. The most essential part of the research is to find the local and global features from the brain cancer images. In this proposed method, both Local Binary Patterns (LBP) and Gray Level Co-occurrence Matrix (GLCM) features are used for feature extraction. The real time brain database is obtained from Jansons MRI Diagnostic centre Erode during November 1, 2013 to December 31, 2014 consisting of 400 images with their ages ranging from 20 to 65 years. In our experiment, 85 samples aretaken for training and the remaining 15 samples are taken for testing. Initially, the local feature information is extracted using LBP method and the overall global features are extracted using GLCM method. By these methods, the brain images are fully illustrated using local and global features. Then the statistical technique is used for feature sub selection where the variance of each features are calculated. The selected features from statistical technique is fed as inputs to the ELM Neural Network classifier where the weights are optimized using Krill Herd algorithm.Results: This proposed hybrid approach achieves 98.9% accuracy when compared with other traditional techniques.