Stroke, primarily categorized as ischemic and hemorrhagic, stands among the top global causes of mortality and can lead to severe neurological conditions if left untreated. The emergence of machine intelligence, especially advanced neural networks, has brought significant advancements in medical imaging, including the automated segmentation of stroke lesions. This automation aids radiologists and surgeons in more effective diagnosis and treatment planning. This review paper concentrates on deep learning methods used in segmenting ischemic and hemorrhagic stroke lesions, examining these models from an architectural standpoint. The paper begins with a discussion on the types of strokes. Subsequently, delve into a review of deep learning models, specifically emphasizing their application in ischemic and hemorrhagic stroke lesion segmentation. The performance, advantages, and limitations of these models are assessed and comparisons among them are presented. The paper concludes by spotlighting the current needs in this field and suggesting potential avenues for future research. The study aims to provide insights to both advanced researchers and newcomers in the field, thereby fostering future advancements in stroke lesion segmentation using deep learning methods.