Abstract: Resting-state electroencephalography (rsEEG) has emerged as a crucial tool in epilepsy research due to its advantages of high temporal resolution and non-invasiveness. Graph theory-based functional connectivity (FC) analysis has revealed common "locally enhanced, globally impaired" characteristics in epilepsy patients, including reduced global efficiency, deviation from small-world properties, and abnormal centrality of key nodes. These topological changes not only facilitate the understanding of the pathological mechanisms of epileptic networks but also assist in seizure detection, epileptogenic focus localization, and treatment outcome prediction. In recent years, studies combining machine learning (ML) and graph neural network (GNN) have further improved the accuracy of rsEEG in seizure prediction and treatment efficacy assessment. However, there are still shortcomings in segmentation strategies, threshold selection, and standardization of analysis procedures. This review summarizes research progress and clinical application prospects based on graph theory, emphasizes its potential value in precise diagnosis and treatment of epilepsy, and proposes that future verification and standardization in large-sample and multi-center studies are necessary.

Key words: Epilepsy, Electroencephalography, Graph theory (not in MeSH), Artificial intelligence, Review

摘要： 静息态脑电图因高时间分辨率和非侵入性优势，成为癫痫研究的重要工具。基于图论的功能连接分析揭示了癫痫患者脑功能网络局部增强、全局受损的特征，包括全局效率下降、小世界特性偏离及关键节点中心性异常，这些拓扑特征不仅有助于理解癫痫网络的病理学机制，还可辅助癫痫发作检测、致痫灶定位及疗效预测。结合机器学习与图神经网络的研究进一步提升静息态脑电图在癫痫发作预测、致痫灶定位和疗效评估中的准确性，但现阶段在频段选择策略、阈值设定及分析流程标准化方面仍存不足。本文综述基于图论的静息态脑电图功能连接在癫痫研究中的进展及临床应用前景，强调其在癫痫精准诊疗中的潜在价值，并提出未来需在大样本多中心研究中验证与规范化。

关键词: 癫痫, 脑电描记术, 图论（非MeSH词）, 人工智能, 综述