Abstract:
Brain tumors are a fatal condition that might result in a very limited life span.
Misdiagnosis can lead to the improper medical action and lower a patient’s prob-
ability of surviving. We address these difficulties by doing extensive experiments
with the proposed framework that uses a hybrid VGG network to classify MRI
slices of healthy brain and brain tumors such as meningioma, glioma, and pituitary.
In this study, we suggest a hybrid deep-learning technique for classifying brain tu-
mors. We used a publicly available brain tumor dataset from Kaggle that included
MRI images of brains with no tumors and three different types of tumors: pituitary,
meningioma, and glioma. On the dataset, preprocessing was done. Our base model
was the VGG16, and we then added CNN layers and dense layers to it. The VGG16
model is trained on the ImageNet dataset. We trained the additional layers that
we added while using transfer learning to freeze the base model layers. A range of
textures, shapes, and features associated with medical imaging are recognizable to
the pre-trained VGG16 model. Moreover, we added extra CNN layers that discov-
ered more complex and distinctive task-related brain tumor pattern features than
the initial ImageNet dataset may have. Our model provided 95.33% test accuracy.
