BRAIN HEMORRHAGE SEGMENTATION ON 3D IMAGES USING U-NET CONVOLUTIONAL NEURAL NETWORK

Abstract

Image segmentation is a main topic in image processing and computer vision with excellent applications such as scene understanding, medical image analysis, robotic perception, video surveillance, augmented reality, and image compression, among many others. Various algorithms for image segmentation have been developed in the literature. The success of U-Net convolutional neural network in clinical image segmentation has been attracting attention. In this thesis, we train the U-Net model on the brain hemorrhage dataset. The data is comprised of Computerized Tomography (CT) scans, 3D images. In particular, a dataset of 82 CT scans of patients with traumatic brain injury. In the end of work, the model based on U-Net achieved 83% accuracy. The current method as stands can be used as an assistive software to the radiologists for the ICH segmentation because it is not yet at a precision that can be used as a standalone segmentation method. The future work can include collecting further CT scans to improve the precision and reliability.