Comparative Analysis of Multiple Pretrained CNN Models To Identify Human Brain Tumor Through Deep Learning.

Abstract

The tumor of brain is a severe disease condition caused by uncontrolled and improper cell division. One of the most deadly and severe cancers that can affect both adults and children is the brain tumor. Timely and precise diagnosis of the tumors presence in brain can help in treatment process. Identifying the brain tumors is among the major essential and tough responsibilities in the field of handling medical image because physical identification with human assistance may result in incorrect prognosis and diagnosis. The use of Computer Aided Diagnosis with the aim to detect tumors of brain has gained attention due to recent advancements in the technology. The health sector has benefited from deep learning application for the diagnosis of numerous disorders regarding medical images. Our research study's goal is to use MRI scans to identify tumors of brain using a deep neural network approach. Convolutional Neural Networks (CNN), a type of deep learning network model, are used in the diagnosis process. We will use a variety of CNN model designs in this research, including ordinary CNN, InceptionV3, MobileNetV2, DenseNet201, ResNet50, ResNet101, VGG19. Scarcity of data is a fact, so we collected brain tumor MRI image data from Kaggle and merged different image files. We applied data augmentation to enhance the amount of MRI data since deep learning networks perform better when they are trained on a big amount of data. A comparative analysis of the accuracy of different models were done to come in a conclusion about which model provides the greatest accuracy in detecting brain tumors from MRI images. DenseNet201 with 99% accuracy has outperformed the accuracy among the evaluated pretrained methods. It will help the physicians to early detect Brain Tumor more precisely using DenseNet201 architecture.