Deep Learning for Stroke Segmentation in Brain Imaging: A Review

Stroke is a medical emergency in which the flow of blood to the brain is decreased due to an interruption caused either by a clot or a burst. It is ranked among the primary reasons of death and a major cause of disability worldwide. Neuroimaging is widely used to assist the diagnosis and treatment of this kind of brain related diseases, particularly Magnetic resonance imaging (MRI) and Computed tomography (CT) scans. Analyzing the brain scans to accurately segment the lesion area and identify the type of the stroke is a crucial step that needs to be done rapidly and precisely. However, it is still a complex and lengthy process due to the variability of the stroke appearances, and the limited number of neurologists able to locate them. In this paper, we explore how deep learning is used for accurate automatic segmentation of ischemic and hemorrhagic strokes in brain scans. We discuss some state-of-the-art approaches based on U-net model proposed by recent studies to emphasize the role of convolutional neural networks for semantic segmentation in medical imaging, and we take an attempt to highlight the impact of multimodality scans on the advancement of stroke treatment.