Brain Tumor Detection Using Transfer Learning in Deep Learning

Abstract

Background/Objectives: Magnetic resonance imaging (MRI) is widely used for tumor evaluation. However, MRI generates enormous data, making manual segmentation difficult in a reasonable amount of time, which limits the use of accurate measurements in clinical practice. Therefore, this study focuses on the automatic and reliable segmentation methods which are needed for early diagnosis of brain tumors. Methods: In this study, we used deep learning-based convolutional neural networks (CNNs) to extract features and automatically classify brain tumors based on MRI images. In addition to conventional CNNs, the application of transfer learning was investigated by using three types of CNNs (Inception-V3, VGG-16, and VGG-19) to achieve reasonable accuracy, with fine tuning of the final layers to improve the accuracy of the models. Findings: The results show that applying transfer learning to a CNN achieves high accuracy in less time and with a smaller dataset. VGG- 19 achieves 97 % accuracy, and VGG-16 achieves 96 % accuracy, which is better than the accuracy of Inception-V3 (89 %). Our proposed model using CNNs with transfer learning provides more robust automatic and reliable segmentation methods. Novelty: The novelty of this work is the use of Transfer Learning in conjunction with deep learning-based CNNs, as Transfer Learning provides a novel technique to analyse data with few annotations by transferring information from the source domain to the target domain. Keywords: Brain tumor detection; Convolutional Neural Networks; Malignant; Benign; Transfer Learning VGGNET; InceptionV3; MRI