基于随机分块模型的结构脑网络连接评价的研究

发布时间：2018-06-06 01:07

本文选题：磁共振弥散加权成像 + 纤维示踪　；参考：《太原理工大学》2017年硕士论文

【摘要】：随着活体磁共振弥散加权成像宏观连接组映射的快速发展,当务之急是根据这些重构数据得到连接强度的衡量指标,并从网络自身角度验证这些连接的可靠性。在猴脑的宏观连接领域,重构纤维束数量更多的被用来作为皮层区域间连接强度的指标,并且通过统计分析验证基于DWI的纤维束数量与动物活体的示踪解剖连接强度之间存在正相关的关系。近年来,磁共振弥散加权成像得到了广泛的应用,但是仍然有一些问题尚未得到解决,比如,获取到的影像数据存在容积效应、纤维交叉的问题。在纤维追踪方法方面,也存在一定的问题,比如参数设置的问题等。由于扫描技术以及追踪技术的问题使得根据纤维方法构造得到的脑网络中存在一定数量的假阳性连接以及假阴性连接;还有,对于长距离连接,通过现有的技术总是很难捕捉到的,而这些问题会导致网络结构有很大的变化,使得依据该网络结构所得到的结论失去真实性。对于非人类的哺乳动物来说,可以得到真实网络,来对比构造网络的正确性,若无法获取得到真实的网络结构(比如人脑),则构造得到的网络中的连接是否可靠以及可靠程度是多少,这方面的研究是比较少的。在本文中,对DWI的派生指标与示踪灌注强度的指标进行了比较,并使用随机分块模型对DWI派生出的连接进行了评价,对评价结果与灌注强度的指标,以及dwi得到连接的其他衡量指标进行了比较。区域连接的连接强度的示踪信息是从两个通用的猴连接组数据集中获得,包括(1)cocomac数据库,收集了猴脑的示踪实验数据,(2)高分辨率的示踪数据集,是由markov和kennedy以及他们的合作伙伴提供的。网络中的数据表示的是重构的纤维路径的纤维束数量,是从23只猴子获得的dwi数据所得到的,对得到的23个个体网络使用符号检验得到中枢网络,从而,对中枢网络中的连接进行评价,结果发现,连接的可靠性值与连接强度的指标是正相关的(p值都小于0.05),与基于dwi得到的连接的其他衡量指标间同样存在强烈相关。在该结论的基础上,将解决重构网络中存在的假阳性连接与假阴性连接的问题,优化网络的结构。本文的主要创新工作有以下几点:第一,基于随机分块模型网络连接可信度计算。按照随机分块模型的思想,将网络中的节点随机划分为相同或不同的组,节点之间的连接是否可靠以及可靠的程度主要依赖与节点所存在的组。通过将连接的可靠性值与连接的真实强度进行相关性分析,结果表明基于随机分块模型得到的连接可靠性与真实强度存在强烈的正相关。第二,结构脑网络连接可靠性验证。将计算所得到的可靠性的值与连接的其他指标进行相关性分析,指标包括纤维束数量,各向异性值,距离值,平均体素个数,结果表明,纤维束数量,各向异性值与平均体素数量这3个指标均与连接的可靠性存在强烈的正相关,距离值指标与可靠性存在强烈的负相关,即基于随机分块模型的方法可以适用于脑结构网络的研究。第三,结构脑网络连接优化。使用成像技术得到的网络中存在一定的假阳性连接与假阴性连接,在本文中使用连接可靠性的值以及符号检验的值对网络中的这两类连接进行优化。结果表明,使用可靠性的值优化后的网络与真实网络更接近,可靠性值可以用于脑结构网络的优化。本文提出基于随机分块模型对DWI的网络连接进行评价的方法,可以正确的计算连接的可靠性,并以此来优化网络结构。实验的结果显示,基于随机分块模型的评价值对白质连接的灌注强度以及连接的其他指标提供近乎真实的估计。本文通过对两种模式的交叉比较提出了一个观点:基于随机分块模型的方法可以作为连接评价的有效方法论。
[Abstract]:With the rapid development of the macroscopical connection group mapping of DWI, the urgent task is to obtain the measurement index of connection strength based on these reconstructed data and verify the reliability of these connections from the network itself. The number of reconstructed fiber bundles is used more as the cortical interconnect in the macaque brain connection field. There is a positive correlation between the number of fiber bundles based on DWI and the intensity of the tracer anatomical connection of animal living body through statistical analysis. In recent years, magnetic resonance diffusion weighted imaging has been widely used, but there are still some problems that have not been solved. For example, the image data obtained are available. There are some problems in fiber tracking, such as the problem of parameter setting, etc., such as the problem of parameter setting, and so on. Due to the problems of scanning and tracking technology, there are a number of false positive connections and false negative connections in the brain network constructed according to the fiber method, and the long distance connection, It is difficult to capture by existing technology, and these problems can lead to a great change in the network structure, which makes the conclusion of the network structure lose authenticity. For non human mammals, the real network can be obtained to compare the correctness of the structure network, if the real network node can not be obtained. In this paper, the derivation index of DWI and the index of tracer perfusion intensity are compared, and a random partitioned model is used to evaluate the connections derived from DWI, and the results and perfusion of the evaluation are given in this paper. The strength index, as well as the other measurements that DWI gets connected, is compared. The tracing information for the connection strength of the region is obtained from two common monkey connection group data sets, including (1) CoCoMac database, collecting the monkey brain tracer experimental data, and (2) high resolution tracer data sets, which are Markov and Kennedy, and they The data in the network represent the number of fiber bundles of the reconstructed fiber path, which is obtained from the DWI data obtained by 23 monkeys. The central network is obtained by using symbolic tests for the obtained 23 individual networks, so that the connections in the central network are evaluated, and the results are found, and the connection reliability values and connections are found. The index of intensity is positively correlated (P value is less than 0.05), and there is a strong correlation with other metrics based on DWI based connections. On the basis of this conclusion, the problem of false positive connections and false negative connections in the reconfigurable network will be solved and the network structure is optimized. The main innovations of this paper are as follows: first, Based on the stochastic block model network connection reliability calculation. According to the idea of random block model, the nodes in the network are randomly divided into the same or different groups. The reliability and reliability of the connections between nodes depend mainly on the groups existing in the nodes. The results show that there is a strong positive correlation between the reliability and the true strength of the connection based on the random block model. Second, the reliability verification of the structural brain network connection. The correlation analysis of the calculated reliability values and the other indexes of the connection is carried out, including the number of fiber bundles, the anisotropic values, the distance values, and the average. The number of voxels shows that there is a strong positive correlation between the number of fiber bundles, the 3 indexes of the anisotropic value and the average volume of body element, and there is a strong negative correlation between the distance value index and the reliability. That is, the method based on the random block model can be applied to the study of the brain structure network. Third, the structure of the brain network is superior to the structural network. There are certain false positive connections and false negative connections in the network using imaging technology. In this paper, the values of connection reliability and the value of symbol test are used to optimize the two types of connections in the network. The results show that the network with the reliability value optimized is closer to the real network, and the reliability value can be used in the brain. In this paper, a method of evaluating the network connection of DWI based on random block model is proposed, which can correctly calculate the reliability of the connection and optimize the network structure. The experimental results show that the evaluation value based on the random partitioned model provides close proximity to the perfusion intensity of white connection and other indicators of connection. Real estimation. This paper puts forward a point of view through the cross comparison of two models: a method based on random block model can be used as an effective methodology for connection evaluation.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R338;O157.5

【参考文献】