Modified CNN for Multi-class Brain Tumor Classification in MR Images with Blurred Edges

Brain tumors are caused by the hyperproduction of specific cells. Only sixty percent of patients with a malignant brain tumor survive for more than a year. As there is no treatment for tumor, the strategy implemented to provide relief to the patient is preventing tumor cells from spreading through chemotherapy or surgery. Only early stages of the cancer can be successfully treated with these approaches, making early cancer detection crucial. However, the doctors with exceptional domain knowledge need not be occupied with this task as it is linear enough to program computers to achieve adequate success in predictions, and hence, the doctors can be involved only when there is an uncertainty in the computer predictions. In order to determine the type of tumor on Magnetic Resonance Images (MRI), this article suggests a novel strategy that combines deep learning and image transformation techniques using Fourier transformation, and classifies brain tumors into four categories, including gliomas, meningiomas, pituitary tumors and no-tumor. The results show that, in comparison to the existing deep learning models, proposed strategy yields a 5% gain in accuracy simultaneously classifying and providing results more quickly.