MRI Image Analysis for Brain Tumor Detection Using Convolutional Neural Network

Abstract

Brain tumors are a dangerous type of cancer with one of the lowest chances of being alive after five years. Magnetic resonance imaging (MRI) is frequently used by neurologists to determine the type of brain tumor present. Using computer-aided tools can speed up the diagnosis process and minimize the burden on health care systems. Deep learning for medical imaging has demonstrated outstanding achievements, especially in the automatic and fast diagnosis of many malignancies. However, in order to get decent results with deep learning models, we require a good quality of data images. To improve the quality of MRI images, a three-step preprocessing method is proposed in this research, coupled with a new Deep Convolutional Neural Network (DCNN) architecture. Batch normalization is used in the design to speed up training and increase the learning rate while also simplifying the initialization of the layer weights. With a modest number of convolutional, max-pooling layers and training iterations, the suggested architecture is computationally light. The proposed architecture is compared to other models in this paper to show its effectiveness. When tested on a dataset of 3394 MRI images, the system achieved a remarkable competitive accuracy. The proposed architecture has been shown to be strong and has helped improve the accuracy of detecting a wide range of brain diseases in a short amount of time.