Multimodal fusion of brain imaging data combines multiple neuroimaging modalities, such as structural and functional MRI, diffusion tensor imaging, and positron emission tomography, to extract complementary and shared information for better understanding of brain structure and function. This review discusses advanced machine learning methods for multimodal fusion, including supervised and unsupervised learning strategies, and their applications in brain atlas construction, cognition, development, and brain disorders. The review highlights the importance of multimodal fusion in improving the prediction of behavioral phenotypes, brain aging, and biomarker discovery for brain diseases. It also discusses emerging trends and challenges, such as handling multi-scale and big data, and the need for new models and platforms. The review covers four main topics: methodologies for multimodal fusion, brain atlasing via multimodal imaging, multimodal fusion in studying cognition and development, and multimodal fusion in brain disorders. It discusses various fusion methods, including correlation-based, clustering-based, and data reconstruction-based approaches, as well as deep learning and graph neural networks. The review also addresses challenges in multimodal fusion, such as big data, multi-scale data, and the need for new models. The review emphasizes the importance of multimodal fusion in clinical applications, including diagnosis, prognosis, and treatment response prediction for brain disorders. The review concludes with a discussion of future directions, including the need for large-scale data, multi-scale data integration, and the development of new models and platforms for brain imaging fusion.Multimodal fusion of brain imaging data combines multiple neuroimaging modalities, such as structural and functional MRI, diffusion tensor imaging, and positron emission tomography, to extract complementary and shared information for better understanding of brain structure and function. This review discusses advanced machine learning methods for multimodal fusion, including supervised and unsupervised learning strategies, and their applications in brain atlas construction, cognition, development, and brain disorders. The review highlights the importance of multimodal fusion in improving the prediction of behavioral phenotypes, brain aging, and biomarker discovery for brain diseases. It also discusses emerging trends and challenges, such as handling multi-scale and big data, and the need for new models and platforms. The review covers four main topics: methodologies for multimodal fusion, brain atlasing via multimodal imaging, multimodal fusion in studying cognition and development, and multimodal fusion in brain disorders. It discusses various fusion methods, including correlation-based, clustering-based, and data reconstruction-based approaches, as well as deep learning and graph neural networks. The review also addresses challenges in multimodal fusion, such as big data, multi-scale data, and the need for new models. The review emphasizes the importance of multimodal fusion in clinical applications, including diagnosis, prognosis, and treatment response prediction for brain disorders. The review concludes with a discussion of future directions, including the need for large-scale data, multi-scale data integration, and the development of new models and platforms for brain imaging fusion.