A CNN-Based Strategy to Classify MRI-Based Brain Tumors Using Deep Convolutional Network

Abstract

Brain tumor is a severe health condition that kills many lives every year, and several of those casualties are from rural areas. However, the technology to diagnose brain tumors at an early stage is not as efficient as expected. Therefore, we sought to create a reliable system that can help medical professionals to identify brain tumors. Although several studies are being conducted on this issue, we attempted to establish a much more efficient and error-free classification method, which is trained with a comparatively substantial number of real datasets rather than augmented data. Using a modified VGG-16 (Visual Geometry Group) architecture on 10,153 MRI (Magnetic Resonance Imaging) images with 3 different classes (Glioma, Meningioma, and Pituitary), the network performs significantly well. It achieved a precision of 99.4% for Glioma, 96.7% for Meningioma, and 100% for Pituitary, with an overall accuracy of 99.5%. It also attained better results than several other existing CNN architectures and state-of-the-art work.

Keywords

• brain tumor
• MRI
• multiclass classification
• deep learning
• VGG
• CNN