Exploring Transfer Learning in Brain Tumor Classification with Deep Neural Networks

In today's era, medical advancements have reached their peak but still, some bugs led to the emergence and involvement of advanced techniques like Machine learning, Artificial Intelligence, and Deep Learning. These three are interconnected and have strong relationships that work as such, to increase the growth of work in any domain with rapid accuracy and without any error. This paper deals with the identification and categorization of tumors. Brain tumors are a result of abnormal behavior of cell division which leads to uncontrolled growth of cells and results in the formation of masses and lumps in the brain. The detection of tumors is done using radiology with the help of MRI techniques or MRI scans. Traditional scans sometimes lead to errors causing misdiagnoses. This process is also timeconsuming and as such requires high expertise. To elevate this process, in this paper, we used a multi-layered CNN model which takes the MRI images as input and classifies it into one of the four major categories i.e., 'Glioma', 'Meningioma', 'Pituitary' or 'No Tumor'. The publicly available dataset used to train this model contains over 7000 images which have been segregated into testing and training where both classes contain 4 different segmentations depicting the different brain tumors. The training folder contains nearly 5200 images, and the testing folder has over 1500 images. The accuracy of the model has been tested to over 96 percent. Then we converted the above model to the VGG16 model by using the study of Transfer Learning and the model gave up to 98.04% accuracy and validation accuracy was approximately 94.73%.