Parallel Convolutional Neural Network for Brain Tumor Using MR Images

We propose a method that utilizes an Auto-Context Convolutional Neural Network (CNN) to learn important local and global image features from 2D patches of varying window sizes. Brain imaging is a crucial first step in many neuroimaging techniques. The precision and efficiency of brain extraction are important to the overall accuracy. Brain extraction algorithms rely on brain atlases that combine questions about anatomy and assume geometry. However, when these assumptions are violated, the effectiveness of this process is limited. As a result, this study presents a CNN-based automated technique for developing an accurate brain extraction tool. We use two different methods to learn local and global images from two dimension patches: 1) Three parallel 2D convolutions based on three different dimensions (axial, coronal and sagittal). This uses a voxel method of implicitly learning 3D data without computationally expensive 3D convolution. 2 U-net architecture-based advanced convolutional network. We employ network post-processing outputs as interactive data, together with raw pictures, to extract local images and brain connections from non-brain tissue. The study proposes a new model for NLM filtering of MRI images. The proposed technique was tested MRI data. We have achieved outstanding results in brain extraction with our CNN algorithm. It has demonstrated better performance than two recently released public reference sets, LPBA40 and OASIS, with Dice overlap coefficients. Our automated content algorithms have enabled us to make significant improvements. We have also evaluated our algorithm's performance in extracting fetal brains from reconstructed MRI datasets with arbitrary orientation Our CNN surpasses previous approaches in extracting fetal brains from MRI. Our algorithm has achieved a Dice coefficient of 95.97%. Iteratively using posterior probability maps and image patches to extract the brain from non-brain tissue.