An Efficient deep learning approach to detect Brain Tumor using MRI images

Abstract

A brain tumor is the development of mutated cells in the human brain. Many di er- ent types of brain tumors exist nowadays. According to researchers and physicians, some brain tumors are non-cancerous while some are life-threatening. In most cases, the cancer is detected at the last stage and it is di cult to recover. This increases the mortality rate. If this could be detected in the initial stages, then a lot more lives could be saved. Nowadays, brain tumors are being detected through auto- mated processes using arti cial intelligence algorithms and brain image data. In this research, we propose an e cient approach to detect brain tumors using Mag- netic Resonance Imaging (MRI) data and a deep neural network. The proposed system comprises several steps - preprocessing and classi cation of brain MRI im- ages. Furthermore, we analyzed the performances of di erent deep neural network architectures and optimized them with an e cient one. The proposed model en- ables classifying brain tumors e ectively with higher accuracy. To commence, we collected data and classi ed it using the ResNetl0l, ResNet50, InceptionV3, VGG19, and VGG19 architectures. As a consequence of our analysis, we obtained an accu- racy rate of 96.72% for VGG16, 96.17% for ResNet50, and 95.55% for InceptionV3. Then, using these three top classi ers, we constructed an ensemble model and ob- tained an overall accuracy rate of 98.60% using EBTDM (Explainable Brain Tumor Detection Model).