Recent advances in medical image segmentation by using deep learning algorithms: a brief review

Medical image segmentation plays a vital role in computer-aided diagnosis (CAD), enabling precise extraction of anatomical and pathological structures to enhance clinical decision-making. The review systematically examines recent advancements in deep learning-based segmentation, with a focus on four key architectures: U-Net variants, Transformer-based models, the emerging Mamba framework, and the Segment Anything Model (SAM). We analyze innovative strategies such as attention mechanisms, multi-scale feature fusion, and lightweight designs that address limitations in computational efficiency, feature representation, and generalization. U-Net-based models demonstrate robustness in handling local features, while Transformers excel at capturing global dependencies. Mamba introduces efficient long-range modeling, and SAM revolutionizes adaptability through prompt-driven segmentation. The review discussed the challenges faced in medical image segmentation, including high computational costs, reliance on annotated data, and domain adaptation for multimodal imaging. By understanding these challenges, we can confirm the future research should prioritize hybrid architectures, self-supervised learning, and optimized training paradigms to improve scalability and clinical applicability. So next-generation segmentation models can further bridge the gap between technical innovation and practical healthcare solutions, ultimately enhancing diagnostic accuracy and patient outcomes.