Deep Transfer Learning Based Multi-Class Brain Tumors Classification Using MRI Images
Mrinmoy Mondal, Md. Farukuzzaman Faruk, Nasif Raihan, Protiva Ahammed
Abstract
A brain tumor is a severe disease that can be fatal and significantly impacts one’s quality of life. The traditional method of identifying tumors relies on physicians, which is time-consuming and prone to errors, putting the patient’s life in jeopardy. Identifying the classes of brain tumors is difficult due to the high anatomical and spatial diversity of the brain tumor’s surrounding region. An automated and precise diagnosis approach is required to treat this severe disease effectively. Deep learning technology, such as CNN, can be used to diagnose various tumor types in the early stages of their development using brain MRI. In this study, a deep transfer learning framework based on VGG-19 is introduced to accurately detect three common kinds of tumors from brain MRI. There are primarily two phases to the suggested framework. The VGG-19 frozen part is the first phase, while the modified neural style classification part is the second phase. With certain modified techniques, the class imbalance impact within the MRI dataset and the generalization error issue during the training process were also resolved. The proposed model has a 94% classification accuracy and a 94% F1-score.