Detection & Classification of MRI Brain Tumor using Convolutional Neural Network
Everyone knows about the importance of healthiness and controlling the diseases. Among all other diseases, cancer considered to be one of the leading causes of morbidity and mortality worldwide. Screening can locate most of the cancers at an early, treatable stage. While research has come a long way in detecting various forms of cancer, there are still many that don't get detected until it's too late. Since the brain is the control room of the body, cancer located in this area can affect many different areas of the body. Diagnosing a brain tumor can be a complicated process and involve a number of specialists. A brain scan, most often an MRI, is the first step. Errors in cancer diagnosis (origins in the clinic setting, diagnostic test or procedure performance, pathologic confirmation of diagnosis, follow-up of patient or test result, or patient-related delays) are likely the most harmful and expensive types of diagnostic errors. Computer-aided detection, are systems that assist doctors in the interpretation of medical images. These systems process digital images for typical appearances and to highlight conspicuous sections, such as possible diseases, in order to offer input to support a decision taken by the professional. In this thesis an efficient algorithm for classifying MRI images using convolutional neural networks is proposed. The performance of the proposed algorithm have shown a great success in classifying large amount of brain tumor MRI images. Classification was applied on 3 groups. Binary classification was performed to distinguish each two classes. Validation accuracies was 100%, 98.4%, 96.9%, 92.2% and 92.2% for normal/tumor, grade2/grade4, grade3/grade4, grade2/grade3 and lower/high grade classes respectively. For 3 classes, grade2/grade3/grade4 and normal/lower grade/high grade, we achieved 85.9% and 93.8% accuracy on validation set. In order to evaluate the network, we compared all the normal, grade2, grade 3 and grade 4 tumors and it results in 85.9% accuracy. The proposed model performs well on MRI images and it could be used to assist doctors in diagnosing and grading Gliomas.