脑肿瘤是由脑组织细胞异常生长引起的，对人类生命构成重大威胁。磁共振成像是一种典型的非侵入式成像技术,它可以产生高分辨率、无损伤和无颅骨伪影的脑影像。随着生物医学技术的不断发展，利用磁共振成像（MRI）技术进行脑肿瘤的诊断和治疗已成为提高患者生存率和降低计算成本的主要技术手段。目的本文对深度学习在多模态MRI脑肿瘤分割领域的最新研究进展和成果进行了系统性的回顾。首先，介绍了多模态MRI脑肿瘤图像分割所涉及的评价指标和公开数据集。其次,对基于U-Net、Transformer、SAM在处理多模态脑肿瘤图像分割的能力上进行了深入探讨，总结了这些技术的优势、局限性，并对各类模型性能进行比较。最后，讨论了当前多模态MRI脑肿瘤图像分割面临的问题与挑战，同时对未来的研究方向进行了展望。意义：本文通过分析不同深度学习模型在多模态MRI脑肿瘤影像上的表现，为医学影像分析领域的研究人员和临床医生提供借鉴，提高脑肿瘤诊断的准确性和效率。

Brain tumors are caused by abnormal growth of brain tissue cells and pose a significant threat to human life. Magnetic resonance imaging is a typical non-invasive imaging technique that can generate high-resolution, non-invasive, and skull artifact free brain images. With the continuous development of biomedical technology, the use of magnetic resonance imaging (MRI) technology for the diagnosis and treatment of brain tumors has become the main technical means to improve patient survival rates and reduce computational costs. Objective: This article systematically reviews the latest research progress and achievements of deep learning in multimodal MRI brain tumor segmentation. Firstly, the evaluation criteria and publicly available datasets involved in multimodal MRI brain tumor image segmentation were introduced. Secondly, an in-depth exploration was conducted on the ability of processing multimodal brain tumor image segmentation based on U-Net, Transformer, and SAM. The advantages and limitations of these technologies were summarized, and the performance of various models was compared. Finally, the problems and challenges faced by current multimodal MRI brain tumor image segmentation were discussed, and future research directions were also discussed. Contribution: This article analyzes the performance of different deep learning models on multimodal MRI brain tumor images, providing reference for researchers and clinical doctors in the field of medical image analysis, and improving the accuracy and efficiency of brain tumor diagnosis.