Study on Data Augmentation Method for Brain Tumour Detection

Abstract

Brain tumours are very dangerous diseases in the world. Brain tumours will cause people to lose their brain function and the survival rate of having brain tumours is low. The characteristics of brain tumours are very complex, so, the accuracy of manual classification to classify benign and malignant brain tumours comes with low accuracy. In this case, the classification method with the use of machine learning (ML) becomes significant since ML classification gets higher accuracy than manual classification. Among ML models, Convolutional Neural Networks (CNN) are widely used to classify brain tumour pictures. In ML, strong models have more parameters and more data are required to train those parameters. However, datasets may not be big enough to train the model. Therefore, few-shot learning methods are proposed to increase the accuracy of the model. In this study, three different methods of data augmentation are used to augment 3,000 pictures’ brain tumour datasets. The codes are successfully run, and the graph of accuracy is exhibited. In this article, the result of the study shows that the cutout method implemented in data augmentation performs the best in detecting brain tumours.