부산대학교 도서관

Brain Tumor is a deadly disease that starts with the abnormal growth of brain cells. It is important to diagnose it early. There are diverse ways like CT scan, Ultrasound, MRI, etc., but MRI (magnetic resonance imaging) images are most widely used because of their less radiation and less ionization. The methods used in this paper are deep learning techniques like MLP (Multi-Layer Perceptron), CNN (convolutional neural networks), and FCNN (fully convolutional neural networks). The dataset used in this paper is BRATS 2018. Performance Metrics of Dice score, PPV (positive predictive value), and sensitivity are used to find the best method among these techniques according to their scores. The results of FCNN have more dice score values compared to CNN and MLP. So, FCNN integrated with CRF (conditional random fields) is an effective technique for detecting brain tumor.