Evaluation of K-Means and CNN Architectures for Segmentation of Brain Tumor

Abstract

A Brain that is accumulated with anomalistic group of cells is tumorand this tumor can be present in any part of the brain. Tumors are consolidated and are of distinct shapes and sizes. Tumors can be fatal and non-fatal.Non-fatal tumors are basically known as primary brain tumors which is originated in your brain, technically known as Benign. Fatal tumors are known as secondary brain tumors which occurs due to deadly cells that diffuse/disperse into the brain which is scientifically called as metastatic brain tumors. There are different techniques for brain tumor segmentation, Deep Learning is one among them and it provides better results when compared to other techniques like Fuzzy Clustering, SVM Technique, Region Growing Method and so on. This article provides/presents K-Means, CNN (LeNet, UNet) architectures to segment brain tumor using MRI images. Kaggle datasets has been used for our work which comprises of multiple images. This proposed network shows the comparision of different architectures and disclose the best architecture that is accurate.

Keywords:Kaggle dataset, Deep Learning, K-Means, CNN, UNet, LeNet