基于群智能算法的脑效应连接网络学习方法研究

Identifying brain effective connectivity networks from functional magnetic resonance imaging (fMRI) data is an important advanced subject in neuroinformatics in recent years, where the research which employs bayesian network learning methods to identify brain effective connectivity networks has become a new hot topic in the field. However, most of the research efforts adopt the greedy search method, and there is almost not intensive research on the search principle. In particular, these existing works ignored the influence of the greedy search method on identifying results. Therefore, this project will employ swarm intelligence optimization mechanisms, and focus on how to improve the accuracy and efficiency, which are two key scientific issues, to perform the following researches: 1) Based on the idea of multi-source information fusion, the project merges the information of multimodal magnetic resonance imaging to establish swarm intelligent search models to improve the quality of brain effective connectivity network learning. 2) Reusing the meta- heuristic information generated by swarm intelligence searches, the project develops some novel approaches to calculate the connectivity strength in learning brain effective connectivity networks. 3) By means of the information exchange and co-evolution between different populations, the project studies the parallel search framework of swarm intelligence, the goal of the research is to develop some new learning methods with high efficiency. 4) Combined with time series characteristics of fMRI data, the project explores new algorithms to learn brain effective connectivity networks using swarm intelligence to parallelly search for dynamic bayesian networks. 5) The project takes Alzheimer's disease (AD) as a typical case for validation and application, reveals properly the abnormal connectivity mode and its evolution law, and provides scientific evidences for AD’s early diagnosis and disease prevention interventions. This proposal focuses on some new attempts and explorations about swarm intelligent methods for learning brain effective connectivity networks from fMRI data. The research will facilitate the development of swarm intelligence theory and its application in brain science, thus has a very important theoretical significance and application value.

脑效应连接网络识别是近年来神经信息学中的一项前沿课题，基于贝叶斯网学习的方法是新兴起的一个研究热点。但目前多数工作采用贪婪搜索方法，缺乏在搜索机理上深入细致的研究，更忽视了贪婪搜索对识别结果的影响。为此，本项目利用群智能全局优化机理，围绕如何提高学习精度和效率两个关键科学问题开展如下研究：基于多源信息融合思想，建立多模态磁共振成像信息融合的群智能搜索模型以提升脑效应连接网络的学习质量；复用群智能搜索产生的元启发信息，探索在脑效应连接网络学习中连接强度计算的新方法；利用种群间协同优化机理，研究群智能并行搜索框架，创建高效率的学习手段；结合时序特征，探索群智能并行搜索动态贝叶斯网的脑效应连接网络学习新算法；以阿尔兹海默病为典型案例进行验证和应用，准确揭示异常连接模式及其演变规律，为疾病早期诊断和预防干预提供科学依据。项目将推动群智能理论的发展及其在脑科学中的应用，具有重要的理论意义和应用价值。

脑效应连接网络识别是近年来神经信息学中的一项前沿课题，基于贝叶斯网学习的方法是新兴起的一个研究热点。针对基于贪婪搜索的学习方法存在的不足，本项目开展了如下几方面的研究：1)对基于功能磁共振成像的人脑效应连接网络识别方法进行了研究综述；2) 利用群智能机理从多模态数据融合的角度入手，探索了基于群智能算法提升脑效应连接网络学习质量的新途径；3) 利用种群间协同优化机理，研究了基于多种群的并行搜索框架，提升了脑效应连接网络的学习效率；4) 结合fMRI数据的时序特征，探索了搜索动态贝叶斯网的脑效应连接网络学习的新算法；5）以阿尔兹海默病为典型案例进行验证和应用，能够揭示脑疾病异常效应连接模式及其演变规律，为疾病早期诊断和预防干预提供了辅助依据；6）拓展了基于深度学习的脑网络分类及疾病诊断的新方法；总之,项目综述了该领域的研究，探索了脑效应连接网络学习、脑网络分类的一些新方法。取得的主要研究成果包括,在国际知名刊物《IEEE Transactions on Image Processing》（CCF A类）、《Pattern Recognition》（CCF B类）、《IEEE Journal of Biomedical and Health Informatics》（CCF C类）、《Frontiers in Neuroscience》、《IEEE/ACM Transactions on Computational Biology and Bioinformatics》（CCF B类）、《Soft Computing》（CCF C类）、《Applied Intelligence》（CCF C类）、人工智能领域旗舰会议AAAI2020（CCF A类）、BIBM2018、2019（CCF B类）、国内期刊《自动化学报》（CCF A类）、《计算机研究与发展》（CCF A类）等上发表论文33篇，其中，SCI期刊论文10篇，包括 CCF推荐的优秀国际刊物9篇（2篇顶级刊物），CCF推荐的A 类中文期刊5篇,申请专利和软著13项。这些研究推动了群智能理论的发展及其在脑效应连接网络学习、脑网络划分等生物信息学中的应用，具有重要的理论意义和应用价值。