Abstract
When we talk about cancer, the most dangerous cancer that is caused by Malignant brain tumors are “Brain Cancers”. It is caused by an uncontrollable growth of a cell in the brain tissue. Recent studies on this brain cancer have shown to have a high mortality rate if not diagnosed in the early stages when the tumor is first detected. So, the main aim of the radiologist and a neurologist is to find the early brain tumor. In order to do that, we need systems that can detect the tumor automatically for the diagnosis. The common systems that medical professionals use like CAD (Computer Aided Diagnostic), MR (Magnetic Resonance) Imaging are shown to have positive outcomes, but accurate tumor detection and classification is still very challenging till date. It is due to the fact that the tumor comes with different structures, proportions, and positions.
Existing Machine Learning models for detection and classification have outstanding accuracy scores. For example, classification models like NasNet, ResNet, and InceptionV3 have accuracy scores of 99.6%, 99.7%, and 97.66%, respectively. But when it comes to the standard classification model like SVM has a very low accuracy of 91% compared to other models. This is due to the fact that it includes drawbacks such as no proper parameter tuning, restricted focus on Data Augmentation, and susceptibility to technological problems.
In this project, we are going to look into an SVM model that can detect the tumor and then classify according to its properties into 4 classes which are Glioma, Meningioma, Pituitary and No-Tumor with a higher accuracy score. Then compare them to other classification models like Naïve Bayes, KNN and Random Forest to see if the SVM model can be a reliable method for classification.