Brain MRI classification for tumor detection with deep pre-trained models

Abstract

Detecting brain tumors at an early stage can improve the likelihood of therapeutic outcome, reduce diagnostic time, and boost the cure rate. Due to its excellent resolution, magnetic resonance imaging (MRI) remains a popular method for detecting brain tumors, is the base for many deep-learning systems that assist clinicians in diagnosing brain cancer early. The purpose of this paper is to evaluate four deep convolutional neural networks with transfer learning models to classify brain cancers. 2518 MRI images used in this study are from two publicly accessible sources. This study aims to investigate the use of pre-trained models (Xception, InceptionV3, MobileNet, and VGG19) to classify brain MRI images with data augmentation preprocessing technique into “tumor” and “no-tumor” classes. For all CNN models’ evaluation, MobileNet outperforms Xception, InceptionV3, and VGG19 with 99.41% accuracy, 99% precision, 99.44% specificity, and 99.38% sensitivity. The Xception model was next with 98.10% accuracy, 98% precision, 97.22% specificity, and 99.07% sensitivity. The InceptionV3 and VGG19 models presented an accuracy of 97.36% and 93.12%, respectively.