虽然意识的概念尚存在争议，但从意识角度理解功能性脑疾病的重要性已经得到了广泛的认可，50多位科学家在《Nature Human Behaviour》发表的《Opportunities and challenges for a maturing science of consciousness》（2019年，3卷，104-107页）文章建议：研究不同功能性脑疾病的意识异常，能更有效地指导临床诊断和干预。目前，在意识障碍、麻醉和睡眠中，人们对意识水平变化及其异常有了充分的理解，但对抑郁、精神分裂等精神疾病的意识异常还有待开展深入研究。

本论文发展了TMS-EEG技术，开展了意识异常脑电活动的特征分析和分类研究。分析了五组不同意识状态受试者（正常、抑郁症、精神分裂症、微意识状态、无反应觉醒综合征）TMS-EEG的TEP和变异性等特征，建立了意识异常的分类模型并对其进行可解释性分析，最后，基于TEP及其小波变换方法，分析了意识异常患者TMS-EEG时-频-空域高维空间的特征及分布。本论文的主要研究及发现如下：

(1) 针对意识异常患者脑电活动的时频特征及分布问题，提取了受试者TMS-EEG的TEP成分，并对其进行溯源、GMFA分析和压榨小波变换，结果显示，不同程度的意识异常患者TEP成分的数量、波幅、潜伏期、空间分布及其激活强度存在差异，五组受试者中央区和顶区诱发Gamma、High Beta和低频带（<13Hz）振荡大小排序为：精神分裂症>抑郁症\健康对照>微意识状态>无反应觉醒综合征。此研究验证了TMS-EEG时频域特征可以用来反映不同意识异常患者大脑活动之间的差异。

(2) 针对意识异常患者脑电活动的稳定性问题，分析了受试者TMS-EEG信号的时域和时频域变异性，结果显示，TMS诱发的神经信号变异性与大脑的意识状态有关；五组受试者时域变异性大小排序为：微意识状态\无反应觉醒综合征>精神分裂症>抑郁症\健康对照；在时频域中，TMS-EEG信号变异性的持续时间随着频率的增加而减小，五组受试者的大小排序与频带范围(Gamma、High Beta、Low Beta)有关。此研究首次揭示了TMS-EEG的神经活动变异性与意识状态相关，为建立意识异常的分类模型提供了更灵敏的脑电特征。

(3) 针对意识异常的分类识别问题，基于深度学习技术和TMS-EEG特征构建了意识异常分类识别模型并对其进行可解释性分析，该模型通过特征学习网络和特征融合网络实现了高阶意识特征的提取和不同维度特征的融合，其总体正确率达到了88.62%，表现出较好的分类性能；分类模型的积分梯度和层电导分析结果发现模型输入特征和隐藏层神经元的权重在意识异常患者的分类识别中存在差异，保证了模型应用的可靠性。

(4) 针对意识异常脑电活动特征的高维空间分布问题，基于张量分解提取了受试者TMS-EEG在时-频-空域中的特异性成分。基于TEP构成的时-空高阶矩阵分析结果显示，不同意识异常患者脑电的振幅在特定时间和脑区增高；基于压榨小波变换的构成时-频-空高阶矩阵分析结果显示，抑郁症、精神分裂症以及微意识状态患者Beta频带振荡的异常增高与大脑对TMS刺激的反应时间和反应区域有关，无反应觉醒综合征患者的低频振荡增高。此研究揭示了TMS-EEG的时-频-空域特征分布在不同的意识异常患者之间存在差异。

综上所述，TMS-EEG的时频空域特征及其高维空间联合分布，可以反映意识异常脑电活动的差异，建立的分类模型在不同意识异常患者的识别中具有较好的表现，其可解释性分析保证了模型应用的可靠性。本论文成果为意识异常的临床诊断和干预提供了重要的技术和方法。

Although the concept of consciousness is still controversial, the importance of understanding functional brain diseases from the perspective of consciousness has been widely recognized. More than 50 scientists suggested that studying the abnormal consciousness of different functional brain diseases can more effectively guide clinical diagnosis and intervention in the paper "Opportunities and challenges for a maturing science of consciousness" which was published in the "Nature Human Behaviour" journal (2019, Volume 3, 104-107 Page). At present, people have full understanded the change of consciousness level and abnormal consciousness in the research about disorder of consciousness, anesthesia and sleep. But the we should further study the abnormal consciousness in mental diseases, such as depression and schizophrenia.

In this paper, we developed TMS-EEG technology, and done EEG feature analysis and classification of subjects with abnormal consciousness. First, we analyzed the TEP and variability characteristics of TMS-EEG in five groups of subjects with different states of consciousness, which including healthy control, depression, schizophrenia, minimally conscious state, unresponsive wakefulness syndrome patients. Then we established an assessment model of abnormal consciousness and performed interpretability analysis on this model. Finally, based on the TEP and its wavelet transform, we explored the characteristics and distribution of TMS-EEG in a high-dimensional space, which contains two or three of temporal, frequency, spatial space, in patients with abnormal consciousness.

The main research and findings of this paper are as follows:

(1) Aiming at the temporal, frequency characteristics and distribution of brain electrical activity in patients with abnormal consciousness, we extracted the TEP components of TMS-EEG data, and done the source location, GMFA and squeezing wavelet transform analysis of TEP. The results showed that patients with abnormal consciousness had different abnormalities in the number, amplitude, latency, spatial distribution and activation intensity of TEP components. The magnitude of TMS evoked oscillations, which was in Gamma, High Beta and low frequency (<13Hz), in the five groups subjects’ central and parietal region was ranked as: schizophrenia > depression \ healthy controls > minimally conscious state > unresponsive wakefulness syndrome. This study verifies that the temporal, frequency features of TMS-EEG can be used to reflect the differences of brain activity between patients with consciousness abnormalities.

(2) Aiming at the stability of brain electrical activity in patients with abnormal consciousness, we analyzed the temporal and temporal-frequency variability of the TMS-EEG data. The results showed that the neural signal variability, which was induced by TMS, was associated with the brain state of consciousness; the magnitude of temporal variability in five groups subjects was ranked as: minimally conscious state \ unresponsive wakefulness syndrome > schizophrenia > depression \ healthy control. In the temporal-frequency space, the duration of TMS-EEG signal variability decreased with frequency increasing, and the rank order of the five groups subjects was related to the frequency band (e.g. Gamma, High Beta, Low Beta). This study firstly revealed that the neural variability of TMS-EEG is related to the state of consciousness, and provides more sensitive EEG features for establishing the classification model of abnormal consciousness.

(3) Aiming at the problem of classification and recognition of patients with abnormal consciousness, based on deep learning technology and TMS-EEG features, we established an assessment model of abnormal consciousness and performed interpretability analysis on this model. Through feature learning network and feature fusion network, this model realized the extraction of high-order consciousness features and the fusion of features in different dimensions. It had a good performance of classifications with the accuracy reached at 88.62%. The results of Integrated Gradients and Layer Conductance of the classification model, which ensures the reliability of the model application, showed that the weights of model input features and hidden layer neurons were different in the classification and recognition of patients with abnormal consciousness.

(4) Aiming at the high-dimensional spatial distribution of EEG activity in patients with abnormal consciousness, we extracted the specific components from subjects' TMS-EEG in the temporal-frequency-spatial space by using tensor decomposition. The analysis of the temporal-frequency high-order matrix, which was constructed by TEP, showed that the amplitude of EEG activity in patients with abnormal consciousness increased at specific times and brain regions. The analysis of temporal-frequency-spatial high-order matrix, which was constructed by squeezing wavelet transform, showed that abnormal increases of Beta band oscillations in depression, schizophrenia, minimally conscious state patients were correlated with the time and region of brain’s response to TMS stimulation, and the low frequency oscillations of patients with unresponsive wakefulness syndrome increased. This study revealed that the distribution of temporal-frequency-spatial characteristics of TMS-EEG were different among patients with abnormal consciousness.

To sum up, the temporal, frequency, spatial features of TMS-EEG and its joint distribution in high-dimensional space can reflect the differences of EEG activity in patients with abnormal consciousness. The assessment model of abnormal consciousness has good performance in the identification of patients with different abnormal consciousness. And its Interpretability analysis ensures the reliability in the application of model. The results of this paper provided important techniques and methods for the clinical diagnosis and intervention of patients with abnormal consciousness.