CAPSULE NETWORKS: HOW EFFECTIVE IT IN BRAIN TUMOR DETECTION

Brain tumor is a critical disease and its accurate and timely detection is important for effective treatment. Several Deep Learning methods are present today to deal with and diagnose the disease on time. CNN is one of the most used Deep Learning techniques for image classification and it has significantly witnessed success in medical image processing, the main shortcoming of CNN is its inability to visualize and analyze spatial hierarchies. CapsNets are used to overcome this disadvantage. It gives more reliable classification results by including Dynamic Routing in place of MaxPooling layers, which is the main reason for the translation invariance in CNN. Dynamic Routing is the brain of CapsNet. The paper uses a public MRI dataset. The proposed model achieves an accuracy of 95.5% and outperforms DL techniques like VGG16, ResNet, and AlexNet. The main performance matrices that are used by this study are Accuracy, Precision, Recall, ROC-AUC. As the comparisons shows that the proposed technique is performing best out of the given techniques.