CLASSIFICATION OF BRAIN TUMORS ON MRI IMAGES USING DEEP LEARNING ARCHITECTURES

Abstract

A brain tumor is a dangerous neural illness produced by the strict growth of prison cells in the brain or head. The segmentation, analysis, and separation of unclean tumor parts from Magnetic Resonance Imaging (MRI) images are the main sources of anxiety. To report the segmented MRI images including tumor, the usage of computer-assisted methods is necessary. In this paper, a Convolutional Neural Network (CNN) approach is applied to identify brain cancers in MRI images. Two datasets are used in this study, namely Kaggle Brain MRI database and Figshare Brain MRI database. Models of deep CNN, consisting of VGG16, AlexNet, and ResNet, are utilized to extract deep features. The classification accuracies of the aforementioned Deep Learning (DL) networks are used to measure the efficiencies of the implemented systems. For the Kaggle database, AlexNet achieves 98% accuracy, VGG16 has 97% accuracy and ResNet has 66% accuracy. Among these networks, AlexNet has provided the highest level of accuracy. In the Figshare dataset, AlexNet and VGG16 both achieve 99% accuracy, and ResNet has 96% accuracy. In terms of accuracy, AlexNet and VGG16 outperform ResNet. These performances aid in the early detection of cancers before they cause physical harm such as paralysis and other complications.