Brain Tumour—Augmentation, Segmentation and Classification Using Deep Learning—A Review

Tumour can be detected early and prevented, although this is not always practicable. Image augmentation and segmentation is an important method used to enhance the properties and abilities of deep learning architectures and can be generalised with the regularisation of the image data. This method plays an important role where the number of original training image data is limited and deriving new attributes from image data becomes expensive and time-consuming. This is a general and common issue in medical image analytics, especially when it is about brain tumour classification and prediction. In this paper, we reviewed the recent enhancement in the field of cancer image generation algorithms used over a number of magnetic resonance brain tumour images. For more understanding of the practical and real aspects of most of the algorithms, our work investigates the articles written and submitted on challenges faced for multimodal brain tumour segmentation. This review also verifies which image augmentation techniques were exploited and what were the research impacts on the capabilities with supervised learning scenarios. In the end, we highlighted the use of pre-trained CNN-based architectural methods such as H2NF, GoogleNet, UNet, etc., in order to serialise and synthesise high-quality automated brain tumour examples which can give a boost to the capabilities of deep learning models.