HUMAN INTELLIGENCE BASED DEEP LEARNING TECHNIQUE FOR IMAGE SEGMENATION OF BRAIN MRI

Abstract

In this work, a fully automated system for brain region segmentation by using Human intelligence based deep learning technique is proposed. Deep learning technique is most popular state of the art method in recent applications. There are two stages involved the pre-processing and segmentation via Convolutional Neural Network (CNN).The MRI image with noise is used as an input image. MRI images are collected from publicly available database Open Access Series of Image Studies (OASIS). Three layers are used in this network, which is used to segment the brain region. The MR images are first given to pre-processing step to enhance the quality of image for segmentation. In this work, Non Local Mean Filter is used for image denoising which calculates weighted average of pixels and finding similarity with the target pixel. The denoised image is given as an input of CNN. Brain region segmentation by deep learning involves feature extraction. CNN learns features directly from an image and no handcrafted features are needed. The method consists of three steps such as input data generation, construction of model and learning the parameter. So, a compact representation from the image as image patches are given as input data to the multilayer convolutional neural network. The supervised deep network consists of three layers. Input image is given to the input layer, it predict the label from input layer. In every hidden layer one convolutional layer and one pooling layer is Present. Convolutional layer compute a dot product of the weights, input, and add a bias term. In this work 4 training images and 1 testing images in ages from the database are used. CNN is trained iteratively with representative input patterns along with target label. The execution of the CNN gives high exactness in the scope of 94% to 96%.