Hypergraph Neural Networks for Brain Tumor Analysis and Medical Image Understanding: A Review

Abstract

Medical image analysis plays a critical role in modern clinical diagnosis and treatment planning. While deep learning, particularly convolutional neural networks, has achieved remarkable success, conventional models struggle to capture the complex high-order relationships inherent in anatomical structures and pathological patterns. Hypergraph neural networks (HGNNs), which generalize standard graphs by allowing hyperedges to connect multiple nodes, have emerged as a powerful paradigm for modeling such high-order interactions. This review provides a systematic overview of hypergraph learning techniques for medical image analysis. We first introduce the mathematical foundations of hypergraphs and categorize existing hypergraph construction methods into similarity-based, topology-based, and learning-based approaches. We then review state-of-the-art hypergraph neural network architectures, including dynamic hypergraph neural networks, hypergraph transformers, and over-smoothing mitigation strategies. Subsequently, we survey key applications in brain tumor classification, functional brain network analysis, medical image segmentation, and survival prediction. Finally, we discuss open challenges and promising future directions, including multi-modal fusion, model interpretability, computational efficiency, and domain generalization. This review aims to provide researchers with a comprehensive understanding of hypergraph learning for medical imaging and to inspire further advancements in this rapidly evolving field.