Brain Tumor Detection and Classification Using Hybrid VGG Network

Abstract

In their highest degree, brain tumors can be extremely deadly. Misdiagnosis can lead to the incorrect course of treatment and lower the likelihood of recovery for patients. To tackle these problems, a hybrid VGG network is extensively researched in the proposed framework to classify brain tumors, including meningioma, pituitary, glioma, and healthy brain MRI slices. In this work, we propose a hybrid model for brain tumor classification. We made use of a publicly available Kaggle brain tumor dataset that included MRI scans showing no brain tumors and three different types of brain tumors: meningiomas, pituitary tumors, and gliomas. We performed preprocessing on the dataset. As our base model, we employed VGG16 and added more CNN layers to it. We trained the additional CNN layers while using transfer learning to freeze the base model layers. The pre-trained VGG16 model is familiar with a variety of textures, forms, and features related to medical imaging. We also added more CNN layers that discovered more complex and distinct task-related brain tumor pattern features that might not have been present in the original ImageNet dataset. We obtained 96.52% test accuracy with our model.