基于脑电时域特征与大脑偏侧化差异表征的抑郁症识别

Depression detection based on EEG temporal-domain features and brain lateralization difference representations

摘要

针对脑电信号时空特征耦合复杂、传统方法难以有效捕捉大脑偏侧化差异与电极关联信息，限制抑郁症诊断精度与可解释性的问题，提出一种基于MLP-Mixer的时-空特征学习网络。首先，通过权重共享的预训练MLP-Mixer学习各EEG通道内的时间上下文信息；随后，以通道级时间特征为输入，利用MLP-Mixer学习大脑左右半球电极对的差异性特征，来表征大脑偏侧化差异与电极间空间关联；最终输出抑郁症二分类诊断结果。在PRED+CT数据集上的实验结果表明，该模型取得96.14%的平均分类准确率。进一步分析显示，α、β双频段特征对抑郁识别的模型性能优于其余频段，与神经生理机制相互佐证，验证该模型特征提取机制的合理性。

Abstract

To address the challenges posed by the complex spatiotemporal coupling of electroencephalogram (EEG) signals, where traditional methods fail to effectively capture lateralization differences and electrode-specific information, which limits the accuracy and interpretability of depression diagnosis, this study proposes a spatiotemporal feature learning network based on MLP-Mixer. A pre-trained MLP-Mixer with shared weights is first adopted to learn temporal context information within each EEG channel. Subsequently, using channel-level temporal features as input, the MLP-Mixer learns differential features between electrode pairs in the left and right hemispheres to characterize lateralization differences and spatial correlations between electrodes. Finally, the model outputs binary classification results for depression diagnosis. Experimental results on the PRED+CT dataset demonstrate that the proposed model achieves an average classification accuracy of 96.14%. Further analysis indicates that dual-bandα andβ frequency features outperform features from other frequency bands in terms of model performance for depression detection. This finding is consistent with neurophysiological mechanisms, thereby validating the rationality of the model's feature extraction mechanism.

关键词

抑郁症 / 脑电图 / 深度学习 / 时-空特征学习

Key words

depression / electroencephalogram / deep learning / spatio-temporal feature learning

参考文献

[1] 刁云恒, 王慧颖, 董娇, 等 . 机器学习在抑郁症辅助诊断中的应用进展[J]. 中国医学物理学杂志, 2022, 39(2): 257-264.

[2] Diao YH, Wang HY, Dong J, et al. Advances in the application of machine learning in auxiliary diagnosis of depression[J]. Chinese Journal of Medical Physics, 2022, 39(2): 257-264.

[3] Shao XX, Sun ST, Li JX, et al. Analysis of functional brain network in MDD based on improved empirical mode decomposition with resting state EEG data[J]. IEEE Trans Neural Syst Rehabil Eng, 2021, 29: 1546-1556.

[4] Khan DM, Masroor K, Jailani MF, et al. Development of wavelet coherence EEG as a biomarker for diagnosis of major depressive disorder[J]. IEEE Sens J, 2022, 22(5): 4315-4325.

[5] 郭朝晖, 王瑜, 马慧鋆, 等 . 基于迁移学习和3D—WGMobileNet的青年抑郁症辅助诊断[J]. 中国医学物理学杂志, 2024, 41(4): 455-462.

[6] Guo ZH, Wang Y, Ma HJ, et al. Diagnosis of youth depression based on transfer learning and 3D—WGMobileNet[J]. Chinese Journal of Medical Physics, 2024, 41(4): 455-462.

[7] Yun S. Advances, challenges, and prospects of electroencephalography—based biomarkers for psychiatric disorders: a narrative review[J]. J Yeungnam Med Sci, 2024, 41(4): 261-268.

[8] 曹轲鸣, 赵璐璐, 赵鸣晖, 等 . 基于穿戴式多生理信号的抑郁智能检测[J]. 中国医学物理学杂志, 2025, 42(9): 1191-1196.

[9] Cao KM, Zhao LL, Zhao MH, et al. Intelligent depression detection based on multi—physiological signals acquired by wearable devices[J]. Chinese Journal of Medical Physics, 2025, 42(9): 1191-1196.

[10] Lin HJ, Fang J, Zhang JP, et al. Resting—state electroencephalogram depression diagnosis based on traditional machine learning and deep learning: a comparative analysis[J]. Sensors (Basel), 2024, 24(21): 6815.

[11] Rehman M, Umar Saeed SM, Khan S, et al. EEG—based depression detection: a temporal domain feature—centric machine learning approach[C]// 2023 International Conference on Frontiers of Information Technology (FIT). Piscataway, NJ, USA: IEEE, 2023: 208-213.

[12] Wu CT, Dillon DG, Hsu HC, et al. Depression detection using relative EEG power induced by emotionally positive images and a conformal kernel support vector machine[J]. Appl Sci, 2018, 8(8): 1244.

[13] Dehghani M, Asayesh V, Torabi Nikjeh M, et al. Investigation capability of EEG—based non—linear features in depression detection[J]. Frontiers Biomed Technol, 2025, 12(4): 722-733.

[14] Xi Y, Chen Y, Meng TY, et al. Depression detection based on the temporal—spatial—frequency feature fusion of EEG[J]. Biomed Signal Process Control, 2025, 100: 106930.

[15] Liu XY, Zhang HR, Cui Y, et al. EEG—based major depressive disorder recognition by neural oscillation and asymmetry[J]. Front Neurosci, 2024, 18: 1362111.

[16] 刘雷, 周仁来 . 一个测量抑郁症的重要神经指标: 静息额叶脑电活动的不对称性[J]. 心理科学进展, 2015, 23(6): 1000-1008.

[17] Liu L, Zhou RL . An important neural indicator of measuring depression: the asymmetry of resting frontal activity[J]. Advances in Psychological Science, 2015, 23(6): 1000-1008.

[18] Bruder GE, Stewart JW, Mcgrath PJ . Right brain, left brain in depressive disorders: clinical and theoretical implications of behavioral, electrophysiological and neuroimaging findings[J]. Neurosci Biobehav Rev, 2017, 78: 178-191.

[19] Luo YW, Tang MC, Fan XW . Meta analysis of resting frontal alpha asymmetry as a biomarker of depression[J]. Npj Ment Health Res, 2025, 4(1): 2.

[20] Tolstikhin IO, Houlsby N, Kolesnikov A, et al. Mlp—mixer: an all—mlp architecture for vision[J]. Adv Neural Inf Process Syst, 2021, 34: 24261-24272.

[21] Ba JL, Kiros JR, Hinton GE . Layer normalization[EB/OL]. (2016—07—21). https://arxiv.org/abs/1607.06450.

[22] Hendrycks D, Gimpel K . Gaussian error linear units (GELUs)[EB/OL]. (2023—06—06). https://arxiv.org/abs/1606.08415.

[23] Lee MH, Kwon OY, Kim YJ, et al. EEG dataset and OpenBMI toolbox for three BCI paradigms: an investigation into BCI illiteracy[J]. Gigascience, 2019, 8(5): giz002.

[24] Cavanagh JF, Napolitano A, Wu C, et al. The patient repository for EEG Data + computational tools (PRED+CT)[J]. Front Neuroinform, 2017, 11: 67.

[25] Zhang ZN, Hu YT, Lu JW, et al. Brain network analysis and recognition algorithm for MDD based on class—specific correlation feature selection[J]. Information, 2025, 16(10): 912.

[26] Xu CY, Wang HG, Zhang L, et al. A spatiotemporal fused network considering electrode spatial topology and time—window transition for MDD detection[C]// 2025 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Piscataway, NJ, USA: IEEE, 2025: 6433-6440.

[27] Liu W, Jia KB, Wang ZZ . Graph—based EEG approach for depression prediction: integrating time—frequency complexity and spatial topology[J]. Front Neurosci, 2024, 18: 1367212.

[28] Wang HG, Meng QH, Jin LC, et al. AMGCN—L: an adaptive multi—time—window graph convolutional network with long—short—term memory for depression detection[J]. J Neural Eng, 2023, 20(5): 056038.

[29] Iyortsuun NK, Yeom S, Jhon M, et al. Fb—Siam: a frequency—band—specific Siamese network for eeg—based depression detection using dual connectivity metrics[J]. http://dx.doi.org/10.2139/ssrn.5335984

基金资助

国家自然科学基金(62103299)