毕业生知识库

获得性免疫缺陷综合征 (acquired immune deficiency syndrome, AIDS) 是由人类免疫缺陷病毒 (human immunodeficiency virus, HIV) 引起的一种致死率极高的恶性传染病，也是全球面临的重大公共卫生问题。据统计，约50%的HIV感染者会进一步发展为HIV认知功能障碍。该疾病显著降低了HIV感染者的生活质量，并进一步增加了其死亡风险，因此已成为HIV感染者临床治疗中亟需重点关注和干预的问题。然而，目前HIV认知功能障碍的临床评估主要依赖于心理学量表测试，该方法存在耗时久、专业要求度高的缺点，无法在短时间内进行重复测试，阻碍了HIV认知功能障碍的客观精准评估，有可能延误最佳治疗时间。因此，寻找HIV认知功能障碍潜在的神经生物学标志物是HIV认知评估领域亟待解决的问题。

磁共振影像 (magnetic resonance imaging, MRI) 具备无创和准确性高的优点，能够从多个角度揭示脑功能和结构变化。已有研究表明，HIV认知功能障碍患者在脑结构和功能方面存在异常。然而，前期研究主要集中于单一模态的影像分析，未充分利用不同模态MRI之间的交叉互补信息。同时，HIV认知功能障碍是一种具有退行性特征的疾病，现有研究主要是基于脑影像学特征开展组水平分析或简单的二分类任务，忽略了在个体水平上前瞻性地预测连续的认知能力并采取提前干预策略的重要意义。因此，借助MRI数据，探索与HIV认知功能障碍相关的神经生理基础，并寻找特异性和敏感性较高的潜在生物学标志物，有望辅助HIV认知功能障碍的客观精准评估，对疾病的及时干预和提升临床疗效具有重要意义。

本论文工作将HIV认知功能障碍的多模态脑影像研究作为切入点，基于组水平和个体水平分析，从①大脑功能结构的协同共变、②认知功能的多模态个体化预测、③结合个体差异结构相似性网络的认知预测框架开发三个方面对该疾病进行了深入研究，主要创新性工作归纳如下：

1.在HIV认知功能障碍领域，首次从结构和功能协同共变的角度，利用认知功能损伤评分作为参考信息，采用有监督的多模态融合方法，结合大脑静息态功能、灰质和白质完整性特征，寻找与HIV认知功能损伤相关的多模态脑影像学标志。研究发现HIV认知功能障碍的功能结构共变成分主要位于后顶叶皮层、视觉皮层以及白质的双侧胼胝体。相比于无监督的多模态融合方法，有监督的多模态融合分析能够利用心理学量表测试得到的认知评分作为先验信息，更特异性地挖掘与HIV认知功能损伤相关的多模态共变成分，揭示了与HIV认知功能障碍有关的脑结构和功能异常共变模式，推进了对HIV认知功能障碍潜在的神经基础的理解。

2. 针对现有HIV认知功能障碍的个性化认知水平评估研究主要集中于简单的二分类任务，缺乏独立数据和泛化性验证的问题，本研究联合脑结构连接和功能连接特征，基于102名HIV感染者构建了多模态认知能力的个体化预测模型，并在两个独立数据集上验证了预测模型的泛化性和特异性，探究了具有重要预测功能的脑连接特征。结果表明，基于多模态脑影像特征的认知能力预测准确率显著优于仅使用单一模态的结果，表明不同模态的MRI影像特征中存在有助于评估个体认知障碍的互补性信息。此外，结合多模态影像学特征、人口统计学和HIV临床特征能够进一步显著提高模型的预测精度 (r = 0.61, p = 2.8e-11)，证实了非成像特征与影像特征联合分析的重要意义。同时，该认知水平评估模型具有优良的泛化性能和对HIV感染者的特异性，在异质性较大的88名HIV感染者独立验证数据集上取得较好的预测结果 (r = 0.47, p = 1.3e-5)，但无法推广到人口统计学上匹配的58名HIV阴性被试数据集上 (r < 0.11, p > 0.1)。研究发现，对HIV认知功能障碍的认知能力具有重要预测功能的特征主要与默认网络内部和躯体运动网络内部的结构连接，躯体运动网络内部和皮下核团网络与其他网络间的功能连接相关。本研究表明，联合多模态脑影像特征能够从不同的角度揭示HIV认知功能障碍的脑功能结构基础，推动了HIV认知功能损伤的个性化评估和认知风险预警的建立，对潜在脑影像学标志物的发掘和验证提供了重要的证据。

3. 针对不同的固定脑模板产生不同形态相似性网络和白质结构连接特征预测性能较弱的问题，本研究结合独立成分分析 (independent component analysis, ICA) 和结构相似性网络的优势，利用大数据空间先验ICA方法，首次提出了基于ICA的个体差异结构相似性网络 (individual difference structural similarity network, IDSSN) 的构建方法，并将其与基于ICA提取的功能连接融合，应用于HIV认知功能障碍的个体化预测研究中。多项对比实验表明，相比于脑模板划分的结构网络，本文提出的基于ICA的IDSSN特征能够更精细的刻画个体差异，不仅提高了HIV感染者整体认知评分的预测准确性 (r = 0.62, p = 2.84e-7)，而且在各认知子域上也取得了更优的预测精度。而基于ICA的IDSSN结合ICA功能网络能够进一步提升评估的准确性 (r = 0.69, p = 1.85e-8)。本研究首次提出了基于ICA的IDSSN构建方法，建立了HIV感染者异常大脑模式与认知症状之间的定量映射关系，推动了HIV认知功能障碍的个体化评估框架的建立，有助于更深入地理解相关的神经病理基础。

Acquired immune deficiency syndrome (AIDS) is a highly lethal malignant infectious disease caused by human immunodeficiency virus (HIV) and a major public health problem worldwide. It is estimated that approximately 50% of people living with HIV will develop HIV-associated cognitive impairment. This disease significantly reduces the quality of life of people living with HIV and further increases the risk of death, making it a critical issue in clinical treatment of people living with HIV. However, the clinical evaluation of HIV-associated cognitive impairment mainly relies on time-consuming and highly specialized psychological tests, which cannot be repeated in a short time, hindering the objective and accurate assessment of HIV-associated cognitive impairment, and possibly delaying optimal treatment time. Therefore, the search for potential neurobiological biomarkers of HIV-associated cognitive impairment is an urgent issue in the field of HIV cognitive evaluation.

Magnetic resonance imaging (MRI) has the advantages of non-invasiveness and high accuracy and can reveal brain functional and structural changes from multiple perspectives. Previous studies have shown that HIV-associated cognitive impairment patients have abnormalities in brain structure and functional activity. However, previous studies mainly focused on single-modal imaging analysis and did not fully utilize the cross-complementary information between different modalities of MRI. At the same time, HIV-associated cognitive impairment is a degenerative disease, and existing studies mainly carry out group-level analysis or simple binary classification tasks based on brain imaging features, ignoring the important significance of prospectively predicting continuous cognitive abilities at the individual level and adopting early intervention strategies. Therefore, by using MRI data, exploring the neurophysiological basis related to HIV-associated cognitive impairment and finding highly specific and sensitive potential biological markers may assist in the objective and accurate assessment of cognitive impairment, and have important implications for timely intervention and improving clinical efficacy of the disease.

This thesis focuses on a multimodal neuroimaging study of HIV-associated cognitive impairment. The study employs both group-level and individual-level analyses to investigate the disease from three perspectives: ①the co-variation of functional brain structures, ②multimodal individualized prediction of cognitive function, and ③the development of a cognitive prediction framework that integrates individual differences in structural similarity networks. The main contributions of this work are summarized as follows:

1. For the first time in the field of HIV-associated cognitive impairment, a supervised multimodal fusion method was used from the perspective of the synergistic covariance of structure and function, using cognitive impairment scores as reference information, combined with resting-state functional connectivity, gray and white matter integrity features, to find multimodal brain imaging markers associated with HIV-associated cognitive impairment. The study found that the covariant components of function and structure in HIV-associated cognitive impairment were mainly located in the posterior parietal cortex, visual cortex, and bilateral corpus callosum. Compared with unsupervised multimodal fusion methods, the supervised multimodal fusion analysis can use cognitive evaluation scores obtained from psychological tests as prior information to more specifically mine brain imaging markers associated with HIV-associated cognitive impairment.

2. Current studies on personalized cognitive assessment of HIV-associated cognitive impairment mainly focus on simple binary tasks, lacking independent data and generalizability validation. Therefore, we combined brain structural connectivity and functional connectivity features and built a multimodal personalized prediction model of cognitive ability based on 102 people living with HIV and validated the generalization and specificity of the prediction model on two independent datasets. The study also investigated the brain connectivity features that are essential for predicting cognitive impairments. Results showed that the accuracy of cognitive ability prediction based on multimodal brain imaging features was significantly better than that based on a single modality, indicating that there was complementary information in different MRI imaging features that contributed to individual cognitive assessment. Additionally, combining multimodal imaging features, demographic, and HIV clinical characteristics can significantly improve the prediction accuracy of the model (r = 0.61, p = 2.8e-11), which confirmd the importance of joint analysis of non-imaging features and imaging features. In addition, the cognitive level assessment model had excellent generalization performance and specificity for people living with HIV. It achieved good prediction results (r=0.47, p=1.3e-5) on an independent validation dataset of 88 people living with HIV with high heterogeneity. However, the model could not be generalized to a dataset of 58 demographic characteristics matched HIV-negative subjects (r<0.11, p>0.1). We found that the features with important predictive functions for cognitive ability in HIV-associated cognitive impairment were mainly related to the structural connectivity within the default mode network and somatomotor network, as well as the functional connectivity between the somatomotor network and subcortical networks, and between the somatomotor network and other networks. This study demonstrated that combining multimodal brain imaging features can reveal the neural basis of HIV-associated cognitive impairment from different perspectives, which promoted the personalized assessment of HIV-associated cognitive impairment and the establishment of cognitive risk warnings and provided important evidence for the discovery and validation of potential brain imaging biomarkers.

3. To address the problem of weak predictive performance of structural connectivity features using different brain templates and modal features, this study combined the advantages of independent component analysis (ICA) and structural similarity network, using the spatial prior ICA method of big data. This study proposed a method for constructing ICA-basd individual difference structural similarity network (IDSSN) for the first time, which was integrated with functional connectivity extracted by ICA and applied to personalized prediction of HIV-associated cognitive impairment. Multiple comparative experiments demonstrated that the ICA-basd IDSSN features can more finely characterize individual differences than the structurally network divided by brain templates. This method not only improved the overall predictive accuracy of cognitive scores for people living with HIV (r = 0.62, p = 2.84e-7), but also achieved better prediction accuracy in various cognitive subdomains. Moreover, the combination of individual differences in ICA-basd IDSSN, and functional networks based on ICA further enhanced the accuracy of assessment (r = 0.69, p = 1.85e-8). This study proposed a method for constructing individual differences in ICA-basd IDSSN for the first time, established a quantitative mapping relationship between abnormal brain patterns of people living with HIV and cognitive symptoms, and promoted the establishment of a personalized evaluation framework for HIV-associated cognitive impairment, which helped to deepen our understanding of the relevant neuropathological basis.