Brain Tumor Diagnosis Using Triplanar Multi - Layer Morphological Convolution Neural Network

Brain tumor diagnosis is one of the most significant and challenging problem in medical image processing, as human-assisted manual might lead to an inaccurate prognosis. Moreover, it's a challenging task when there is a significant volume of data to be processed. Since brain tumors can have a variety of appearances and resemble normal tissues, it can be challenging to identify tumor regions in images. Deep learning (DL), particularly convolutional neural networks (CNNs), has seen a substantial growth in application in the field of computer vision and image analysis in recent years. But the traditional convolution fails in the manner of global information loss, where 2D convolution ignores the information on z-axis, 3D convolution networks require heavy computations. DeepLab was built to get efficient work done in the field of neural networks so as to retain all the local and global information about an image. In this paper, we propose a triplanar multi-layer morphological convolution neural network to extract brain tumor from 2D brain Magnetic Resonance Images (MRI) which improves DeepLab method in the aspects of reducing the number of computations and time complexity so as to make the cost-effective system. The proposed system is evaluated based on mIOU and dice score, which shows out proposed methods yields the better performance when compared to state of the art methods.