基于时空图卷积神经网络的脑电信号抑郁识别研究

Research on Depression Recognition of EEG Signals Based on Spatio-Temporal Graph Convolutional Neural Network

摘要: 抑郁症已成为全球广泛流行的神经心理疾病，但传统的诊断手段因主观性影响，常导致较高的误诊率。鉴于此，开发一种更为客观且高效的抑郁症识别方法显得尤为重要。本研究构建了一种基于时空图卷积神经网络(ST-GCN)的脑电信号抑郁分类模型，将时间序列的脑电信号转化为脑拓扑图，捕捉大脑复杂的空间结构信息。同时，引入时空注意力机制，从时间和空间两个维度上有效提取关键信息。具体而言，时空图卷积神经网络结合了空间图卷积和时间卷积的优势，分别用于捕获脑电信号的空间布局特征和时间动态特性。实验结果显示，在公开的脑电数据集HUSM上，该模型的分类准确率、灵敏度以及特异度优于其他基线模型，充分验证了该模型在抑郁症识别方面的优越性能。

Abstract: Depression has become a globally widespread neuropsychiatric disorder, but traditional diagnostic methods, due to their subjectivity, often lead to a high rate of misdiagnosis. In light of this, developing a more objective and efficient depression recognition method is particularly important. This study proposes a depression classification model for electroencephalogram (EEG) signals based on Spatio-Temporal Graph Convolutional Neural Network (ST-GCN). This model converts the time series of EEG signals into brain topological graphs, thereby capturing the complex spatial structure information of the brain. Meanwhile, by introducing a spatio-temporal attention mechanism, the model can effectively extract key information from both temporal and spatial dimensions. Specifically, the ST-GCN combines the advantages of spatial graph convolution and temporal convolution, used to capture the spatial layout features and temporal dynamic characteristics of EEG signals, respectively. Experimental results show that on the public EEG dataset HUSM, the model’s classification accuracy, sensitivity, and specificity are superior to other baseline models, fully verifying the superior performance of this model in depression recognition.