Improved Brain Tumor Segmentation and Diagnosis Using an SVM-Based Classifier

Abstract

The Increased focus on the aspects in the Computer Science Field like Machine Learning and Neural Networks have brought about a change in diagnosis methods of various diseases. Machine Learning has advanced so much and provides benefits like enhanced efficiency and fast decision making which prove to be crucial in the field of medical diagnosis. One of the highly deemed research fields which happen to be greatly affected by this advance is the field of cancer diagnosis. Cancer was responsible for 18 Million cases in 2018 and was responsible for over 9 million deaths. Over the years, the cancer rate has been decreasing marginally, but the last 3 years have seen an increase by 2%. Accurate diagnostic methods need to be considered to when dealing with cancer because incorrect diagnosis can be even fatal to the patient, hence a tool is required, which can trump even the human brain at diagnosing cancer. Machine learning can prove to be this tool due to the fact that it learns along the way once you feed an input to it, so as time goes on it gets smarter and smarter eventually always exceeding the accuracy percentage of its results and thereby proving to be a useful tool for medical diagnosis. In this paper, we propose a support vector machine (SVM) classifier for brain tumor segmentation and diagnosis. 50 MRI images have been used to test this and an accuracy of 100% has been achieved in diagnosis of the tumor as benign or malignant. Kernel accuracy of 90% in each case has also been achieved.