CNN based multi-view classification and ROI segmentation: A survey

In today's world, one of the reasons in rise of mortality among people is cancer. A cancerous disease is bound to occur due to the ungovernable growth of certain cells that can scatter to other parts of the body. The different types of cancerous diseases are lung cancer, breast cancer, brain cancer, skin cancer. One among them which is of major concern is the brain cancer. With the emergence of AI-ML techniques, detection of cancerous tumour can be automated. One of the efficient methods for the detection of brain tumour is convolutional neural network. Visual information from various viewpoints is frequently used by humans in their decision-making process. For the recognition of the brain tumour a single image showing an object is insufficient. Multi-view classification aims to improve classification accuracy by combining data from several perspectives into a uniform comprehensive representation for downstream tasks. To aim that it presents a trustworthy multi-view classification, a classification approach that dynamically integrates diverse perspectives at an evidence level, resulting in a new paradigm for multi-view learning. By incorporating data from each view, the method promotes both classification reliability and resilience by combining several viewpoints. The process of segmenting images involves separating areas within a picture into distinct classes in order to identify them and classify them. In CNN there are different architectures like E-Net, T-Net, W-Net to determine the ROI and perform the image segmentation. In order to automate detection of the brain tumour, MRI image segmentation plays vital role. In this paper, a survey on the various image segmentation approaches and its comparison is presented. The main focus here is on strategies that can be improved and optimized over those that are already in use.