Enhancing VGG-16 by Using Batch Normalization for Classification of Brain Tumors

Abstract

In this paper, we present a study on brain tumor detection using the VGG-16 model, a convolutional neural network known to be effective for computer vision tasks. This study proposes an enhanced VGG-16 model by modifying its architecture with batch normalization and dropout layers to improve generalization and prevent overfitting. Unlike previous studies that directly applied VGG-16, this research introduces architectural changes to adapt the model for more effective multi-class classification of brain tumors. The objective of this study is to accurately determine the presence or absence of a brain tumor by classifying magnetic resonance imaging (MRI) images. The dataset used consists of MRI images of brain tumors classified into two classes: No (no tumor) and YES (tumor). The methodology includes environment setup, data import and preprocessing, VGG16 model construction, and performance evaluation using metrics such as accuracy, precision, and recall. The results show an accuracy of about 98.47% on the validation set and 98.32% on the test set, indicating that the VGG-16 model has the potential to assist medical professionals in diagnosing brain tumors. This study contributes to the field of medical image analysis and provides ideas for applying deep learning to brain tumor diagnosis.