毕业生知识库

精神分裂症是一种高致残性、高异质性的重性精神疾病，严重影响个人的生活状态和社会功能。当前精神分裂症病因尚不明确，诊断仍需依赖于量表评估，缺乏客观的生物标记物，这为医生的临床实践带来了巨大的挑战。磁共振成像技术的发展为开展大规模脑影像研究以刻画精神分裂症的脑结构和功能异常模式提供了可能。然而，现行疾病分类学中关于精神分裂症的二分类诊断严重忽略了其内在的异质性，阻碍了精神分裂症影像学研究的进一步发展。充分尊重精神分裂症的异质性事实，从临床异质性以及影像病理异质性等多个层次开展针对性研究对改善精神分裂症不尽如人意的临床现状具有重要意义。为此，以异质性问题为核心，本文利用多中心大样本数据库基于精细的人类脑网络组图谱从多个角度对精神分裂症的影像学异常模式进行了探究，并尝试通过客观的影像学指标发现其潜在的影像学亚型，为探究精神分裂症的病理机制提供新的见解。本文的主要内容和创新点归纳如下：

1. 基于多中心数据的精神分裂症功能连接异常模式组水平荟萃分析与面向临床异质性的个体化预测研究

多中心大规模影像学数据的采集与分析能够有效增强相关分析的可靠性，但可能会造成因站点差异而导致的数据异质性。为解决这一问题，我们对来自七个中心的精神分裂症影像学数据进行了荟萃分析，其中包含530名健康对照以及531名精神分裂症患者，探究了精神分裂症的全脑功能连接异常模式，在组水平上为精神分裂症的脑网络障碍属性提供了可靠的影像学证据。以个体化预测为目标的精神分裂症辅助诊断模型辅以可解释性技术，能够挖掘个体层次的影像学异常模式，从而对组水平的分析结果进行有效的补充。然而，以简单的二分类任务为导向，忽略了精神分裂症的临床异质性。考虑到精神分裂症的临床症状评分能够在一定程度上反映其临床异质性，我们构建了双空间映射网络，利用症状评分对隐层决策空间中的个体表征进行约束。该模型具备良好的辅助分类与症状评分预测性能，同时能够对潜在的个体化异常功能连接进行挖掘，具有良好的临床探究与应用价值。

2. 基于精神分裂症影像形态学异质性的亚型挖掘

以客观的影像学指标对精神分裂症进行亚型划分是解决其异质性问题的有效方法之一。基于结构磁共振影像数据（来自521名健康对照以及534名精神分裂症患者）提取基于张量的形态学度量，我们发现该影像形态学指标能够捕捉到精神分裂症异质性的影像学异常模式。基于此发现，我们进一步采用稀疏聚类算法将精神分裂症划分为两个稳定的亚型。此外，两类影像形态学异常模式存在差异的亚型在精神分裂症的阴性症状中也存在显著的差异，表明该影像学亚型具备一定的临床相关性。该亚型划分结果为理解精神分裂症的结构异常模式提供了新的见解，为进一步探究精神分裂症影像学异质性与临床异质性的关联提供了基础。

3. 精神分裂症影像学亚型的临床异质性和生物学异质性分析

在利用基于张量的形态学度量发现了两类在阴性症状上存在显著差异的精神分裂症影像学亚型后，我们进一步利用偏最小二乘相关分析建立了两类亚型之间影像学差异与临床异质性之间的关系，明确了两类亚型临床差异的影像学来源，例如小脑区域。该影像-临床关联分析还发现了一个影像异质性与临床异质性共享的隐变量，能够将两类亚型进行有效的关联，为进一步理解两类亚型间的关系提供了新的视角。此外，通过影像-转录关联分析我们探究了两类亚型之间影像学差异所暗示的生物学基础，结果表明精神分裂症在化学突触传递、脑发育等生物过程中可能存在异质性。综上，该部分研究建立了多层次的影像学亚型分析策略，通过数据驱动的方式为精神分裂症在多层次上的异质性研究提供了新的见解。

Schizophrenia is a severe mental disease with high disability and heterogeneity, which seriously affects the living conditions and social functions of patients. At present, the etiology of schizophrenia is still unclear, the diagnosis still relies on the scale evaluation, lacking objective biomarkers, which brings great challenges to clinical practice. The development of magnetic resonance imaging technology makes it possible to carry out large-scale neuroimaging research to depict the brain structure and functional abnormal patterns of schizophrenia. However, the dichotomous diagnosis of schizophrenia in the current nosology seriously ignores its inherent heterogeneity, which hinders the development of neuroimaging-based schizophrenia research. It is of great significance to fully consider the heterogeneity of schizophrenia and carry out targeted research from multiple levels, such as clinical heterogeneity and neuroimaging pathology heterogeneity, to improve the unsatisfactory clinical situation of schizophrenia. Therefore, taking the heterogeneity as the core aspect, this study explores the abnormal neuroimaging patterns of schizophrenia from multiple perspectives based on the fine-scale human brainnetome atlas using a multi-site large sample dataset and attempts to discover potential subtypes through objective neuroimaging indicators, providing new insights into the pathological mechanism of schizophrenia. The main contents and innovations of this paper are summarized as follows:

1. Group-level meta-analysis of schizophrenia abnormal functional connectivity pattern, and individual-level prediction constrained by clinical heterogeneity using multi-site data

The collection and analysis of multi-site large-scale imaging data can effectively enhance the reliability of correlation analysis but may cause data heterogeneity due to site differences. To solve this problem, we conducted a meta-analysis of schizophrenia using the magnetic resonance image (MRI) datasets collected from seven sites, including 530 healthy controls and 531 patients with schizophrenia, and explored the pattern of abnormal brain functional connectivity in schizophrenia, providing reliable group-level imaging evidence for the brain network disorder characteristic of schizophrenia. The computer-aided diagnosis model of schizophrenia targeting individualized prediction, supplemented by interpretable technology, can mine the individual-level abnormal functional connectivity, thus effectively supplementing the analysis results at the group level. However, guided by simple dichotomous tasks, the clinical heterogeneity of schizophrenia was ignored. Considering that the clinical symptom scores of schizophrenia can reflect its clinical heterogeneity to a certain extent, we constructed a dual space mapping network (DSM-Net), using the symptom scores to constrain the individual representations in the hidden decision-making space. The model has good classification and symptom score regression performance, and can also mine the potential individual abnormal functional connectivity, which has clinical exploration and application value.

2. Mining subtypes of schizophrenia based on neuroimaging morphological heterogeneity

Stratifying schizophrenia into subtypes with objective neuroimaging indicators is one of the effective approaches to solving the problem of heterogeneity of schizophrenia. Based on the tensor-based morphological (TBM) measurements extracted from structural MRI of 521 healthy controls and 534 patients with schizophrenia, we found that TBM can capture the heterogeneous neuroimaging abnormal patterns of schizophrenia. Based on this finding, we further clustered schizophrenia into two stable subtypes using sparse clustering. In addition, the two subtypes with different abnormal TBM patterns also have significant differences in the negative symptoms of schizophrenia, indicating that these neuroanatomical subtypes have clinical relevance. The results provide new insights for understanding the structural abnormality pattern of schizophrenia, and provide a basis for further exploring the relationship between neuroimaging heterogeneity and clinical heterogeneity within schizophrenia.

3. Analyses of clinical and biological heterogeneous factors reflected by the neuroanatomical subtypes

After identifying two subtypes of schizophrenia that have significant differences in negative symptoms, we further established the relationship between neuroimaging differences and clinical heterogeneity between the two subtypes using partial least squares correlation analysis, and located the neuroimaging source (one of the most representative regions is the cerebellum) for clinical differences between the two subtypes. The neuroimaging-clinic association analysis also found a latent variable shared by neuroimaging heterogeneity and clinical heterogeneity, which can effectively bridge the two subtypes, providing a new perspective for further understanding the relationship between the two subtypes. In addition, we explored the heterogenous biological factors implied by the neuroimaging differences between the two subtypes through neuroimaging-transcription association analysis. The results showed that schizophrenia may be heterogeneous in biological processes such as “chemical synaptic transmission” and “head development”. To sum up, this part of the study has established a multi-level neuroimaging subtype analysis strategy, which provides new insights into the heterogeneity of schizophrenia at multiple levels through a data-driven approach.