Electroencephalogram (EEG) large models have emerged as a significant research direction in EEG analysis due to their exceptional representation learning capabilities. To systematically review progress in this field, the core theories and key technologies underlying these models are first analyzed, examining their theoretical foundations and design principles. Subsequently, a comprehensive comparison of mainstream EEG large models is conducted across dimensions including model concepts, architectural designs, feature extraction methods, learning paradigms, and model scales. A particular emphasis is placed on comparing and analyzing the binary classification performance of the datasets and downstream tasks used by these models. Finally, based on the analytical results, future development directions for EEG large models are discussed. A knowledge framework of EEG large models is established to serve as a reference for subsequent research and promote in-depth development in the field.
2024年,基于Transformer架构,文献[10]提出脑电大模型LaBraM(Large Brain Model),该模型能够高效捕捉脑电信号的时空特征,进而生成富含语义的神经编码,实现通用表征学习,并在异常检测、事件分类、情绪识别以及步态预测等多个下游任务上,展现出超越现有最优深度学习模型的卓越性能。脑电大模型本质上是一种经过预训练的深度神经网络模型,可以把它比作坚固的基座,充当着各类下游任务的骨干网络。通常情况下,脑电大模型会在规模庞大且具有高度多样性的数据集上开展训练工作,以此来精准捕捉通用的脑电特征。其最为显著的优势体现在强大的表示学习能力上,能够在不同的任务场景之间实现出色的泛化,展现出良好适应性和通用性。经过一年多发展,EEGFormer[11]、EEGPT(EEG Pretrained Transformer)[12-13]、Brant-X(Brain Neural Transformer)[14]、FoME(Foundation Model for EEG)[15]、EEG-GPT[16]、NeuroLM(Neurological Language Model)[17]等多个以Transformer框架为核心架构的脑电大模型相继出现,不仅在算法层面突破了传统模型的性能天花板,更在应用层面为实时脑状态监测、神经疾病筛查等场景提供了技术可行性。
1)EEG-GPT模型。EEG-GPT模型由伊利诺伊大学厄巴纳—香槟分校提出,是一种创新的EEG信号处理框架,利用大型语言模型(Large Language Model, LLM)进行EEG信号的分类和解释[16]。该模型将EEG信号视为一种“外语”,通过LLM的强大能力,不仅实现了高效的分类,还提供了中间推理步骤,显著增强了模型在临床环境中的可解释性和可信度。在仅使用2%的训练数据的情况下,EEG-GPT能够实现与当前最先进的深度学习方法相当的性能,展示了其在数据稀缺环境下的强大适应性。
SHARMAR, MEENAH K. Emerging trends in EEG signal processing: a systematic review[J]. SN Computer Science, 2024,5(4):415.
[2]
BRAKEN, DUC F, ROKOSA, et al. A neurophysiological basis for aperiodic EEG and the background spectral trend[J]. Nature Communications, 2024,15:1514.
[3]
DE GANSC J, BURGERP, VAN DEN ENDEE S, et al. Sleep assessment using EEG-based wearables–A systematic review[J]. Sleep Medicine Reviews, 2024,76:101951.
[4]
RAHULJ, SHARMAD, SHARMAL D, et al. A systematic review of EEG based automated schizophrenia classification through machine learning and deep learning[J]. Frontiers in Human Neuroscience, 2024,18:1347082.
[5]
SEKKALR N, BEREKSI-REGUIGF, RUIZ-FERNANDEZD, et al. Automatic sleep stage classification: from classical machine learning methods to deep learning[J]. Biomedical Signal Processing and Control, 2022,77:No.103751.
[6]
VASWANIA, SHAZEERN, PARMARN, et al. Attention is all you need[J]. Advances in neural information processing systems, 2017,30(1):2.
[7]
MAOW L, FATHURRAHMANH I K, LEEY, et al. EEG dataset classification using CNN method[J]. Journal of Physics: Conference Series, 2020,1456(1):No.012017.
[8]
ROYS, KIRAL-KORNEKI, HARRERS. ChronoNet: a deep recurrent neural network for abnormal EEG identification[C]∥Proceedings of 17th Conference on Artificial Intelligence in Medicine. Cham, Switzerland: Springer, 2019:47-56.
[9]
SONGY H, ZHENGQ Q, LIUB C, et al. EEG conformer: convolutional transformer for EEG decoding and visualization[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2023,31:710-719.
[10]
JIANGW B, ZHAOL M, LUB L. Large brain model for learning generic representations with tremendous EEG data in BCI[DB/OL]. (2024-05-29)[2025-09-24].
[11]
CHENY Q, RENK, SONGK T, et al. EEGFormer: towards transferable and interpretable large-scale EEG foundation model[DB/OL]. (2024-01-11)[2025-09-24].
[12]
WANGG Y, LIUW C, HEY H, et al. EEGPT: pretrained transformer for universal and reliable representation of EEG signals[J]. Advances in Neural Information Processing Systems, 2024,37:39249-39280.
[13]
YUET T, GAOX G, XUES N, et al. EEGPT: unleashing the potential of EEG generalist foundation model by autoregressive pre-training[DB/OL]. (2024-10-14)[2025-09-24].
[14]
ZHANGD Z, YUANZ Z, CHENJ R, et al. Brant-X: a unified physiological signal alignment framework[C]∥Proceedings of 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York, USA: ACM, 2024:4155-4166.
[15]
SHIE Z, ZHAOK, YUANQ L, et al. FoME: a foundation model for EEG using adaptive temporal-lateral attention scaling[DB/OL]. (2024-09-19)[2025-09-24].
[16]
KIMJ W, ALAAA, BERNARDOD. EEG-GPT: exploring capabilities of large language models for EEG classification and interpretation[DB/OL]. (2025-03-20)[2025-09-24].
[17]
JIANGW B, WANGY S, LUB L, et al. NeuroLM: a universal multi-task foundation model for bridging the gap between language and EEG signals[DB/OL]. (2024-08-27)[2025-09-24].
YANGC Q, WESTOVERM B, SUNJ M. BIOT: biosignal transformer for cross-data learning in the wild[J]. Advances in Neural Information Processing Systems, 2023,36:78240-78260.
[20]
ZHANGD Z, YUANZ Z, YANGY, et al. Brant: foundation model for intracranial neural signal[J]. Advances in Neural Information Processing Systems, 2023,36:26304-26321.
[21]
WANGB X, FUX W, LANY, et al. Large transformers are better EEG learners[DB/OL]. (2023-08-20)[2025-09-24].
[22]
DUANY Q, ZHOUJ Z, WANGZ, et al. DeWave: discrete EEG waves encoding for brain dynamics to text translation[DB/OL]. (2023-09-25)[2025-09-24].
[23]
LIUH W, HAJIALIGOLD, ANTONYB, et al. EEG2-TEXT: open vocabulary EEG-to-text decoding with EEG pre-training and multi-view transformer[DB/OL]. (2024-05-03)[2025-09-24].
[24]
WANGZ H L, JIH. Open vocabulary electroencephalography-to-text decoding and zero-shot sentiment classification[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2022,36(5):5350-5358.
[25]
LIUX H, LIUY K, WANGY S, et al. EEG2video: towards decoding dynamic visual perception from EEG signals[J]. Advances in Neural Information Processing Systems, 2024,37:72245-72273.
[26]
LAIJ H, WEIJ Y, YAOL, et al. A simple review of EEG foundation models: datasets, advancements and future perspectives[DB/OL]. (2025-04-24)[2025-09-24].