Efficient Framework for Brain Tumour Classification using Hierarchical Deep Learning Neural Network Classifier
Francis H. Shajin, P. Salini, P. Rajesh, Venu Kadur Nagoji Rao
Abstract
In this manuscript, an efficient framework is proposed for brain tumour classification (BTC) based on hierarchical deep-learning neural network (HieDNN) classifier. Here, the input images are collected from Brats dataset. The input images are preprocessed using Savitzky-Golay denoising method to decrease the noise. The image features, like texture features, are removed using grey-level co-occurrence matrix (GLCM). The extracted images are fed to Hie DNN classifier, which is helpful for classifying the brain images. The proposed system is implemented in MATLAB. The proposed method attains higher accuracy of 31.14%, 16.09% and 11.48% during benign; during malignant higher accuracy 35.18%, 19.17% and 22.80%; during normal higher accuracy 44.20%, 29.97% and 20.44% compared with the existing methods, like convolutional neural network for BTC depending on MRI images (CNN-BTC), microscopic brain tumour detection with classification utilising 3D CNN and feature selection architecture (3DCNN-BTC), BTC utilising hybrid deep auto-encoder using Bayesian fuzzy clustering-based segmentation methodology (DAEN-BTC), respectively.