CASIA OpenIR  > 脑网络组研究
基于脑网络和机器学习的脑动静脉畸形致痫风险研究
廖小华
Subtype硕士
Thesis Advisor宋明
2022-05
Degree Grantor中国科学院自动化研究所
Place of Conferral中国科学院自动化研究所
Degree Name工程硕士
Degree Discipline计算机技术
Keyword脑动静脉畸形 癫痫 脑网络 机器学习
Abstract

 

脑动静脉畸形是一种复杂的脑血管病变,其特点是在病变部位的脑动脉和脑静脉之间缺乏毛细血管,从而形成畸形血管团。对于未破裂脑动静脉畸形患者,癫痫发作比例为24-40%,是其最常见的临床表现。但脑动静脉畸形患者癫痫发作与否的机制仍不明确,难以对脑动静脉畸形患者的癫痫发作风险进行有效预测。因此,本文基于天坛医院收集的197 例未破裂脑动静脉畸形患者的结构磁共振成像数据,以及356 例来自课题组前期收集的正常人静息态功能磁共振成像数据,使用脑网络和机器学习技术,探究未破裂脑动静脉畸形患者癫痫发作背后的神经机制,并建立致痫风险预测模型,为脑动静脉畸形致痫风险研究提供新的发现。

本文假设未破裂脑动静脉畸形患者癫痫发作与病灶位置无关,而是与功能连接的改变相关。本文首先采用基于体素的损伤映射技术来验证这一假设,发现未破裂脑动静脉畸形患者的病变位置是异质的,没有发现与癫痫发作相关的脑区。进一步通过脑网络技术发现,癫痫发作患者的病灶映射到一个共同的功能网络,该网由与楔前叶的负连接和与左尾状核和小脑前叶的正连接定义,并且这种连接模式是脑动静脉畸形相关癫痫发作特异的。基于得到的这一脑网络区域能够对脑动静脉畸形患者的癫痫发作风险进行预测,并在独立验证集中得到验证。通过与7 个典型的静息状态网络叠加分析,本文发现,额顶控制网络、边缘网络和默认模式网络是未破裂脑动静脉畸形患者的致痫网络。

为了进一步验证之前的假设,本文将未破裂脑动静脉畸形患者的结构磁共振像数据和基于正常人计算得到的功能连接数据作为特征,使用机器学习算法对未破裂脑动静脉畸形患者癫痫发作与否进行分类。本文发现,无论是在训练集还是测试集中,对于未破裂脑动静脉畸形患者,使用结构像数据的分类性能显著低于使用功能连接数据的性能。这进一步说明了不同病灶的位置虽然在不同的脑区,但是属于一个共同的脑网络。局部位置的损伤引发了整体脑网络功能的改变,从而诱发了癫痫。本文建立的机器学习最优模型分类AUC 达80%,其在训练集和测试集的准确率为分别为75.46% 和67.65%,说明模型对于未破裂脑动静脉畸形癫痫发作具有较好的预测效果,这将有助于未破裂脑动静脉畸形患者癫痫发作的精确诊断。本文还分析了机器学习模型的可解释性,本文发现基于脑网络的方法所影响的脑区和基于机器学习的方法发现的重要脑区具有一致性,为探究未破裂脑动静脉畸形引发癫痫发作的神经机制提供了新的发现。

总之,本文对未破裂脑动静脉畸形患者致痫风险进行了多方面的研究,发现了与未破裂脑动静脉畸形相关的癫痫网络,建立了具有较高准确率的未破裂脑动静脉畸形癫痫发作预测模型。本文还对脑动静脉畸形患者癫痫发作背后的神经机制进行了解释,为理解疾病机理提供了线索和依据,具有重要的科学意义和潜在的临床价值。

 

Other Abstract

Brain arteriovenous malformation is a complex cerebrovascular disease characterized by the lack of capillaries between the cerebral arteries and cerebral veins at the lesion site, resulting in the formation of malformed vascular masses. For patients with unruptured brain arteriovenous malformation, the proportion of seizures is 24-40%, which is the most common clinical manifestation. However, the mechanism of seizures in patients with brain arteriovenous malformation is still unclear, and it is difficult to effectively predict the risk of seizures in patients with brain arteriovenous malformation. Therefore, based on the structural magnetic resonance imaging data of 197 patients with unruptured brain arteriovenous malformation collected by Tiantan Hospital, and the resting-state functional magnetic resonance imaging data of 356 normal people collected in the early stage of the research group, , this paper used brain network and machine learning technology to explore the neural mechanism behind epileptic seizures in patients with unruptured brain arteriovenous malformation, and established a risk prediction model for epilepsy, which provided new findings for the study of epilepsy risk of brain arteriovenous malformation.

This paper hypothesized that seizures in patients with unruptured brain arteriovenous malformations are not related to the location of the lesion, but rather to changes in functional connectivity. This paper first used a voxel-based lesion mapping technique to test this hypothesis and found that lesions in patients with unruptured brain arteriovenous malformations are heterogeneous in location, with no single region thought to be anatomically associated with seizures. Further through brain network techniques, we found that the lesions of  seizure patients map to a common functional network defined by negative connections to the precuneus and positive connections to the left caudate nucleus and anterior cerebellum, and that this connectivity pattern is specific for brain arteriovenous malformation-related seizures. Based on the obtained brain network region, the seizure risk of patients with brain arteriovenous malformation can be predicted and verified in an independent validation set. By overlaying analysis with seven typical resting state networks, we found that frontoparietal control network, limbic network and default mode network are epileptogenic networks in patients with unruptured brain arteriovenous malformations.

In order to further verify the previous hypothesis, this paper used the structural magnetic resonance imaging  data of patients with unruptured brain arteriovenous malformation and the functional connectivity data calculated based on normal persons as features, using machine learning algorithms to classify seizures in patients with unruptured brain arteriovenous malformations. This paper found that, for patients with unruptured brain arteriovenous malformations, the classification performance using structural image data was significantly lower than that using functional connectivity data, both in the training set and in the testing set.This further indicated that although the locations of different lesions are in different brain regions, they belong to a common brain network. Locally lesion triggers changes in global brain network function that induce epilepsy.The best machine learning model established in this paper has a classification AUC of 80%, and its accuracy in the training set and testing set is 75.46% and 67.65%, respectively, indicating that the model has better performance for unruptured cerebral arteriovenous malformation seizures, which will aid in the precise diagnosis of seizures in patients with unruptured brain arteriovenous malformations.This paper also analyzed the interpretability of machine learning models, and we found that the brain regions affected by the brain network-based method are consistent with the important brain regions discovered by the machine learning-based method, which provided insights into the neural mechanism of epileptic seizures caused by unruptured brain arteriovenous malformations.

In conclusion, this paper conducted a multifaceted study on the risk of epilepsy in patients with unruptured brain arteriovenous malformation, found the epilepsy network associated with unruptured brain arteriovenous malformation, and established a high accuracy prediction model for unruptured brain arteriovenous malformation seizures. This paper also explained the neural mechanism behind epileptic seizures in patients with brain arteriovenous malformations, which provided clues and basis for understanding the disease mechanism. This paper has important scientific significance and potential clinical value.

 

Pages82
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/48719
Collection脑网络组研究
毕业生_硕士学位论文
Recommended Citation
GB/T 7714
廖小华. 基于脑网络和机器学习的脑动静脉畸形致痫风险研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2022.
Files in This Item:
File Name/Size DocType Version Access License
基于脑网络和机器学习的脑动静脉畸形致痫风(9583KB)学位论文 开放获取CC BY-NC-SA
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[廖小华]'s Articles
Baidu academic
Similar articles in Baidu academic
[廖小华]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[廖小华]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.