Abstract:
Brain tumor is a serious life threatening disease. Early diagnosis of disease is very important for treatment. There are more than 120 types of brain tumor. It is very difficult task to classify tumor manually because of small size and complex structure of brain. There is a lot of work to classify brain tumor into normal or abnormal, work for benign and malignant tumor classification is also available but there is very limited work available to classify tumor by its type. Glioblastoma is among the most aggressive types of a brain tumor. There are two modules of our research i.e classification and detection. This report Classify tumor into glioblastoma (GBM) and non glioblastoma (non- GBM) types and also detect the GBM tumor region from MRI images. Deep learning gained a lot of interest in machine learning field over the past few years. We use CNN for classification and R-CNN, faster R-CNN find a tumor region from MRI images. We use transfer learning rather than training a model from scratch because training a model from scratch require a large data set, high computational power and lot of time. We use both the feature extraction and fine tuning techniques to find results. We deployed several pre-trained models those are trained on Image Net. Highest classification rate of 95.40% is achieved by using Vgg16 model. We achieve highest detection precision of 87% by using Alexnet with faster R-CNN.
