第一章 基于多模态磁共振影像的嗅觉障碍患者结构及功能脑网络研究

【摘要】

目的：旨在通过多模态磁共振影像技术，探讨嗅觉障碍（olfactory dysfunction, OD）患者的结构网络、功能网络及其耦合特征，揭示嗅觉障碍的神经机制。

方法：纳入120名嗅觉障碍患者及48名健康对照，通过扩散张量成像（diffusion tensor imaging, DTI）和静息态功能磁共振成像（resting-state functional magnetic resonance imaging, rs-fMRI）构建结构网络和功能网络，结合图论分析和模块化分析方法，评估全局和局部拓扑属性、模块内及模块间连接特征。此外，采用结构-功能耦合（structural-functional coupling, SC-FC）分析方法，计算全脑、脑区和子网络水平的结构-功能耦合值，探讨结构网络与功能网络之间的关系。采用Spearman相关分析，探讨多模态影像特征与嗅觉功能评分、焦虑评分和抑郁评分之间的关系。

结果：在结构网络中，嗅觉障碍患者表现出全局拓扑属性异常，例如路径长度增加、全局效率降低以及层次性减弱。局部拓扑属性分析显示，关键脑区（如前眶额皮层、杏仁核）的节点效率显著升高。模块化分析表明，患者组在躯体运动网络（somatomotor network, SOM）与背侧注意网络（dorsal attention network, DAN）之间的连接增强。在功能网络中，嗅觉障碍患者的网络同配性降低，基底节和丘脑等脑区的节点聚类系数和节点效率显著升高。模块间分析显示，健康对照组在视觉网络（visual network, VIS）与默认模式网络（default mode network, DMN）之间的功能连接显著强于患者组。结构-功能耦合分析发现，嗅觉障碍患者在左侧前眶额皮层、右侧内眶额皮层和右侧后海马等脑区的耦合值显著高于健康对照组。相关性分析表明，嗅觉功能评分与部分脑区及网络的拓扑属性和耦合值存在相关性，例如右侧后海马的耦合值与嗅觉辨别能力呈负相关。

结论：本研究从结构、功能及耦合的多角度揭示了嗅觉障碍患者的脑网络异常，表明嗅觉障碍患者可能通过神经可塑性调节和网络重组代偿嗅觉功能减退。

第二章 基于多模态磁共振影像机器学习的嗅觉减退和嗅觉丧失特征研究

【摘要】

目的：基于多模态磁共振影像数据，采用机器学习算法筛选嗅觉减退与嗅觉丧失患者的关键影像特征，以揭示潜在的神经机制差异。

方法：纳入120名嗅觉障碍患者，其中73名为嗅觉减退，47名为嗅觉丧失。通过高分辨率结构磁共振成像（structural magnetic resonance imaging, sMRI）、弥散张量成像（diffusion tensor imaging, DTI）和静息态功能磁共振成像（resting-state functional MRI, rs-fMRI），提取与嗅觉相关的大脑结构和功能特征，包括大脑皮层厚度、脑区活动、网络连接强度和结构-功能耦合值等。采用Wilcoxon检验和最小绝对值收缩和选择算子（least absolute shrinkage and selection operator, LASSO）回归进行特征筛选，使用8种机器学习算法（包括支持向量机、梯度提升机、神经网络等）构建分类模型，并通过曲线下面积（area under the curve, AUC）、准确率、敏感性、特异性等指标评估模型性能。采用SHAP法（SHapley Additive exPlanations）进行模型可解释性分析。

结果：通过sMRI、DTI和rs-fMRI提取了1930个多模态磁共振影像特征，采用Wilcoxon检验和LASSO回归进行特征筛选，最终筛选出29个关键特征。XGBoost模型在测试集上表现最佳，AUC值为0.755（95% CI：0.593-0.917），灵敏度为71.4%，特异性为76.2%，准确率为74.3%。SHAP分析揭示了左侧额上回皮质厚度、右侧前扣带回的低频振幅波动、fMRI网络的左侧背岛叶皮层的节点局部效率等特征对模型预测的重要贡献。决策曲线分析（decision curve analysis, DCA）表明，XGBoost模型具有较高的临床实用性。

结论：本研究通过机器学习特征选择方法，识别出嗅觉减退与嗅觉丧失患者的显著差异脑区特征，并探讨了这些特征在嗅觉信息处理中的潜在作用，为深入理解嗅觉障碍的神经病理机制提供了新的线索。

第三章 影像转录组学和孟德尔随机化研究嗅觉丧失对大脑皮质结构的影响

【摘要】

目的：旨在通过高分辨率结构磁共振成像、影像转录组学分析和孟德尔随机化方法，探究嗅觉丧失对大脑皮质结构的影响及其潜在的分子机制。

方法：研究纳入52名嗅觉丧失患者和50名健康对照者，使用高分辨率结构磁共振成像（structural magnetic resonance imaging, sMRI）提取大脑皮质表面积（cortical surface area, SA）和皮质厚度（cortical thickness, TH）。选取嗅觉丧失相关遗传变异作为工具变量，结合ENIGMA数据库的皮质SA和TH遗传关联数据，采用孟德尔随机化方法评估嗅觉丧失对皮质结构的因果影响。结合艾伦人脑图谱（Allen Human Brain Atlas, AHBA）的基因表达数据，采用偏最小二乘回归（partial least squares regression, PLSR）和汇总数据的孟德尔随机化（summary-data-based mendelian randomization, SMR）方法，识别嗅觉丧失、SA和TH变化相关的基因，并通过基因富集分析和细胞类型特异性分析揭示其分子机制。

结果：嗅觉丧失患者的左侧前扣带回喙部和颞上沟的SA显著降低（P < 0.05，FDR未校正），而左侧楔叶、缘上回、中央前回等区域的TH显著增加（P < 0.05，FDR校正）。孟德尔随机化分析显示，嗅觉丧失与11种大脑皮层结构特征（如额上回皮层厚度、额中回尾部皮层厚度等）存在因果关联。基因表达分析揭示了与SA和TH变化相关的基因集，其中3460个基因与SA变化相关，1888个基因与TH变化相关。富集分析表明，这些基因显著富集于突触相关通路，如“突触后膜”和“化学突触传递调控”。细胞类型分析显示，SA和TH相关基因在γ-氨基丁酸（Gamma-Aminobutyric Acid , GABA）和谷氨酸神经元中富集。

结论：本研究探讨了嗅觉丧失与大脑皮质结构改变之间的因果关联，揭示了突触功能和神经元可塑性在嗅觉丧失中可能的作用机制。

第四章 嗅觉障碍患者药物及嗅觉训练治疗与tDCS治疗前后脑结构-功能耦合变化研究

【摘要】

目的：旨在探讨药物及嗅觉训练治疗，以及药物及嗅觉训练联合经颅直流电刺激（transcranial direct current stimulation, tDCS）治疗对嗅觉障碍患者嗅觉功能及脑结构-功能耦合的影响，以期为嗅觉障碍的治疗提供新的思路和依据。

方法：研究分为两部分。首先纳入了33名嗅觉障碍患者，接受药物治疗（甲钴胺、鼻喷激素等）及嗅觉训练，持续3-6个月。其次纳入了19名嗅觉障碍患者，在药物治疗及嗅觉训练的基础上联合tDCS治疗，tDCS治疗持续21天，每天30分钟，电流强度为1.5mA，刺激部位为前额叶。在治疗前后，患者接受了多模态磁共振成像（MRI）检查，获取弥散张量成像（diffusion tensor imaging, DTI）和静息态功能磁共振成像（resting-state functional magnetic resonance imaging, rs-fMRI）数据。基于磁共振影像数据，构建结构网络和功能网络，计算不同脑区、子网络的结构-功能耦合（structural-functional coupling, SC-FC）值。通过分析治疗前后SC-FC耦合值的差异，评估其与临床症状的相关性，探讨治疗对脑网络的影响。

结果：药物及嗅觉训练治疗后，患者的嗅觉察觉能力（T）、辨别能力（D）、识别能力（I）及嗅觉总分（TDI）均显著改善（P < 0.05）。脑区水平的SC-FC分析显示，右侧额上回、颞上回、海马旁回等脑区的耦合值发生显著变化。子网络水平的分析表明，背侧注意网络的SC-FC耦合值显著升高，且视觉网络与嗅觉网络、躯体感觉网络与嗅觉网络等功能连接显著增强。药物及嗅觉训练联合tDCS治疗后，患者的嗅觉察觉能力（T）和视觉模拟量表（VAS）评分改善（P < 0.05），但嗅觉辨别能力（D）和识别能力（I）未见显著变化。脑区水平的SC-FC分析显示，右侧额上回、左侧额中回、眶回、中央前回等脑区的耦合值发生显著变化（P < 0.05）。子网络间的功能连接分析表明，视觉网络与边缘网络、躯体感觉网络与腹侧注意网络的功能连接在治疗后增强（P < 0.05）。此外，嗅觉辨别能力（D）和TDI总分与右侧眶回的结构-功能耦合值呈负相关。

结论：药物及嗅觉训练治疗可显著改善嗅觉障碍患者的嗅觉功能，促进脑区与子网络结构-功能耦合发生显著变化，表明神经可塑性增强。联合tDCS治疗在一定程度上改善了患者的嗅觉察觉能力和视觉模拟量表评分，但对于嗅觉辨别和识别能力的提升效果不显著，需进一步深入研究。

Chapter 1: Study on Structural and Functional Brain Networks in Patients with Olfactory Dysfunction Based on Multimodal Magnetic Resonance Imaging

Objective: This study aims to explore the structural and functional networks, as well as their coupling characteristics, in patients with olfactory dysfunction (OD) using multimodal magnetic resonance imaging (MRI) techniques, thereby revealing the neural mechanisms underlying olfactory dysfunction.

Methods: A total of 120 patients with olfactory dysfunction and 48 healthy controls were included. Structural and functional networks were constructed using diffusion tensor imaging (DTI) and resting-state functional magnetic resonance imaging (rs-fMRI). Graph theory and modularity analysis were employed to assess global and local topological properties, as well as intra- and inter-modular connectivity features. Additionally, structural-functional coupling (SC-FC) analysis was conducted to calculate the coupling values at the whole-brain, regional, and subnetwork levels, exploring the relationship between structural and functional networks. Spearman correlation analysis was used to investigate the relationships between multimodal imaging features and olfactory function scores, anxiety scores, and depression scores.

Results: In the structural networks, patients with olfactory dysfunction exhibited abnormal global topological properties, such as increased path length, reduced global efficiency, and weakened hierarchy. Local topological property analysis revealed significantly increased nodal efficiency in key brain regions, including the anterior orbitofrontal cortex and amygdala. Modularity analysis indicated enhanced connectivity between the somatomotor network (SOM) and the dorsal attention network (DAN) in the patient group. In the functional networks, patients with olfactory dysfunction showed decreased network assortativity, with significantly increased nodal clustering coefficients and nodal efficiency in regions such as the basal ganglia and thalamus. Inter-modular analysis revealed that the functional connectivity between the visual network (VIS) and the default mode network (DMN) was significantly stronger in the healthy control group compared to the patient group. Structural-functional coupling analysis demonstrated that patients with olfactory dysfunction had significantly higher coupling values in the left anterior orbitofrontal cortex, right medial orbitofrontal cortex, and right posterior hippocampus compared to healthy controls. Correlation analysis indicated that olfactory function scores were significantly associated with the topological properties and coupling values of certain brain regions and networks, such as a negative correlation between the coupling value of the right posterior hippocampus and olfactory discrimination ability.

Conclusion: This study reveals abnormalities in the brain networks of patients with olfactory dysfunction from structural, functional, and coupling perspectives, suggesting that patients with olfactory dysfunction may compensate for the decline in olfactory function through neural plasticity regulation and network reorganization.

Chapter 2: Machine Learning-Based Feature Analysis of Hyposmia and Anosmia Using Multimodal Magnetic Resonance Imaging

Objective: To identify key neuroimaging features distinguishing hyposmia from anosmia patients using machine learning algorithms on multimodal magnetic resonance imaging (MRI) data, thereby elucidating potential differences in neural mechanisms.

Methods: A total of 120 patients with olfactory dysfunction, including 73 with hyposmia and 47 with anosmia, were included. High-resolution structural magnetic resonance imaging (sMRI), diffusion tensor imaging (DTI), and resting-state functional MRI (rs-fMRI) were used to extract brain structural and functional features related to olfaction, including cortical thickness, regional brain activity, network connectivity strength, and structural-functional coupling values. Feature selection was performed using the Wilcoxon test and least absolute shrinkage and selection operator (LASSO) regression. Eight machine learning algorithms, including support vector machines, gradient boosting machines, and neural networks, were employed to construct classification models. Model performance was evaluated using metrics such as the area under the curve (AUC), accuracy, sensitivity, and specificity. Model interpretability was analyzed using SHapley Additive exPlanations (SHAP).

Results: A total of 1930 multimodal MRI features were extracted from sMRI, DTI, and rs-fMRI. After feature selection using the Wilcoxon test and LASSO regression, 29 key features were identified. The XGBoost model performed best on the test set, with an AUC value of 0.755 (95% CI: 0.593-0.917), sensitivity of 71.4%, specificity of 76.2%, and accuracy of 74.3%. SHAP analysis revealed that features such as cortical thickness in the left superior frontal gyrus, low-frequency amplitude fluctuations in the right anterior cingulate cortex, and nodal local efficiency in the left dorsal insular cortex of the fMRI network contributed significantly to the model's predictions. Decision curve analysis (DCA) indicated that the XGBoost model has high clinical utility.

Conclusion: This study employed machine learning-based feature selection to identify significantly different brain regional features between hyposmia and anosmia patients. The findings provide insights into the potential role of these features in olfactory information processing, offering new clues for understanding the neuropathological mechanisms underlying olfactory dysfunction.

Chapter 3: Investigating the Impact of Anosmia on Brain Cortical Structure Using Imaging Transcriptomics and Mendelian Randomization

Objective: This study aims to investigate the impact of anosmia on brain cortical structure and its underlying molecular mechanisms using high-resolution structural magnetic resonance imaging (sMRI), imaging transcriptomics, and Mendelian randomization.

Methods: The study included 52 patients with anosmia and 50 healthy controls. High-resolution structural MRI (sMRI) was used to extract cortical surface area (SA) and cortical thickness (TH). Genetic variants associated with anosmia were selected as instrumental variables, and Mendelian randomization was performed using genetic association data from the ENIGMA database to assess the causal effects of anosmia on cortical structure. Gene expression data from the Allen Human Brain Atlas (AHBA) were integrated, and partial least squares regression (PLSR) and summary-data-based Mendelian randomization (SMR) were used to identify genes associated with anosmia, SA, and TH changes. Gene enrichment analysis and cell-type-specific analysis were conducted to reveal the molecular mechanisms.

Results: Patients with anosmia showed significantly reduced SA in the left rostral anterior cingulate cortex and superior temporal sulcus (P < 0.05, uncorrected for FDR), while increased TH was observed in the left cuneus, supramarginal gyrus, and precentral gyrus (P < 0.05, FDR-corrected). Mendelian randomization analysis revealed significant causal associations between anosmia and 11 cortical structural features, such as cortical thickness in the superior frontal gyrus and caudal middle frontal gyrus. Gene expression analysis identified 3460 genes associated with SA changes and 1888 genes associated with TH changes. Enrichment analysis showed that these genes were significantly enriched in synaptic-related pathways, such as "postsynaptic membrane" and "regulation of chemical synaptic transmission." Cell-type analysis indicated that SA- and TH-related genes were significantly enriched in GABAergic and glutamatergic neurons.

Conclusion: This study explored the causal relationship between anosmia and changes in brain cortical structure and revealed the potential role of synaptic function and neuronal plasticity in anosmia.

Chapter 4: Changes in Brain Structural-Functional Coupling in Olfactory Dysfunction Patients Before and After Drug and Olfactory Training Therapy Combined with tDCS

Objective: This study aims to investigate the effects of pharmacological and olfactory training treatments, as well as combined pharmacological, olfactory training, and transcranial direct current stimulation (tDCS) treatments, on olfactory function and brain structural-functional coupling in patients with olfactory dysfunction, providing new insights and evidence for the treatment of olfactory dysfunction.

Methods: The study was divided into two parts. First, 33 patients with olfactory dysfunction received pharmacological treatment (methylcobalamin, nasal corticosteroids, etc.) and olfactory training for 3-6 months. Second, 19 patients with olfactory dysfunction received combined pharmacological treatment, olfactory training, and tDCS treatment. tDCS was administered for 21 days, 30 minutes daily, with a current intensity of 1.5 mA, targeting the prefrontal cortex. Before and after treatment, patients underwent multimodal MRI, including diffusion tensor imaging (DTI) and resting-state functional MRI (rs-fMRI). Structural and functional networks were constructed, and structural-functional coupling (SC-FC) values at the regional and subnetwork levels were calculated. Changes in SC-FC values before and after treatment were analyzed, and their correlations with clinical symptoms were assessed to explore the impact of treatment on brain networks.

Results: After pharmacological and olfactory training treatment, patients showed significant improvements in olfactory detection (T), discrimination (D), identification (I), and total olfactory scores (TDI) (P < 0.05). Regional SC-FC analysis revealed significant changes in coupling values in the right superior frontal gyrus, superior temporal gyrus, and parahippocampal gyrus. Subnetwork analysis showed significantly increased SC-FC coupling values in the dorsal attention network, as well as enhanced functional connectivity between the visual and olfactory networks and between the somatosensory and olfactory networks. After combined pharmacological, olfactory training, and tDCS treatment, patients showed significant improvements in olfactory detection (T) and visual analog scale (VAS) scores (P < 0.05), but no significant changes in olfactory discrimination (D) and identification (I). Regional SC-FC analysis revealed significant changes in coupling values in the right superior frontal gyrus, left middle frontal gyrus, orbital gyrus, and precentral gyrus (P < 0.05). Subnetwork analysis showed enhanced functional connectivity between the visual and limbic networks and between the somatosensory and ventral attention networks (P < 0.05). Additionally, olfactory discrimination (D) and TDI scores were negatively correlated with SC-FC values in the right orbital gyrus.

Conclusion: Pharmacological interventions and olfactory training demonstrated significant improvements in olfactory function among patients with olfactory dysfunction, accompanied by marked alterations in structure-function coupling across brain regions and sub-networks, suggesting enhanced neuroplasticity. While combined tDCS therapy showed modest improvements in odor detection thresholds and visual analogue scale (VAS) scores, its effects on odor discrimination and identification capabilities remained statistically non-significant, warranting further investigation.