A Review on Comparison of VGG-16 and DenseNet algorithms for analysing brain tumor in MRI image

Abstract

Brain tumor is a deadly disease for which proper treatment should be given. Brain tumor detection is usually done with the help of analyzing MR images. Accurate brain tumor detection is important for effective treatment. VGG16 and DenseNet are two popular CNN algorithms used widely in medical fields in order to detect diseases from medical images. VGG-16 supports 16 layers. It has three fully connected layers.It is capable of classifying thousand images of thousand different categories. DenseNets are divided into different dense blocks in which there are dense connections between layers through dense blocks. All the layers are connected using matching feature- map sizes and it works in a feed-forward nature. Each layers obtain additional inputs from all preceding layers and passes its own feature-maps to all subsequent layers. The proposed system has five major stages, namely, image pre-processing, image enhancement, image segmentation, brain tumor image classifica- tion using VGG-16 and brain tumor image classification using DenseNet. This work compares the performances of both vgg- 16 and DenseNet algorithms and find out which algorithm gives better
accuracy in detection. Pre-trained vgg-16 and DenseNet models are used and around 4000 MRI scan images are used for testing and training.