Performance Analysis of Deep Learning Techniques in Brain Tumor Segmentation

Although there has been rapid development in brain tumor segmentation, it still remains one of the most challenging tasks in medical image analysis. Manual image segmentation techniques are time consuming and prone to human error. The main objective of automated tumor segmentation is to provide an accurate delineation of the tumor area. Recently, Deep Neural Networks (DNN) have provided excellent results in computer vision problems. DNN techniques can assist physicians to accurately detect brain tumors in early and advanced stages of growth. Moreover, precise segmentation can support medical practitioners in diagnosis and at formulating an appropriate treatment plan. In this study we present the segmentation of two-dimensional (2D) brain tumor images using Magnetic Resonance Imaging (MRI) images. Although the U-Net model was specifically designed for medical image segmentation, this study shows that the VGG-16 and ResNet-50 models’ performance is comparable and as effective as the U-Net model. Segmentation analysis is performed for each baseline model and compared using the Kaggle datasets of brain tumor gliomas.