基于静息态fMRI多功能属性建模的快速脑功能指纹识别研究

Brain functional fingerprints (BFF) have essentially scientific value in many aspects such as revealing the individual intrinsic functional structure, mapping the individual brain atlas, diagnosing the brain diseases very early, etc. However, the BFF recognition using the functional magnetic resonance imaging technology has been currently trapped by many deficiencies, such as the inaccuracy of brain activity decoding, poor discrimination of functional features, low anti-variability of the identification classifier, slow speed of fingerprints recognition, etc. Thus, in this project, the theory, models and algorithms of fast BFF recognition are studied. First, the multifunctional attributes constrained functional connectivity detection model is proposed to make sure of decoding brain activity accurately, by modeling multiple functional properties of the human brain as a whole, i.e., functional separation, functional integration, parsimony of functional connection, sparse coding and multiresponse of the cortex, which is quite beneficial to capture the commonness and individual specificity of brain functional networks (BFN). Secondly, the multi-domain feature extraction method with respect to the BFNs is studied, which extracts kinds of functional features from the time, spatial and functional connectome domain, respectively, fully describing the individual intrinsic functional structure. Thirdly, the brain functional fingerprints selection and classification models based on multi-domain features fusion are proposed, which first select the key functional fingerprints automatically, then accurately identify the brain fingerprints. Finally, the parallel processing algorithms regarding BFF recognition process are developed to accelerate the identification of brain fingerprints. The proposed models and methods of BFF identification in this project, can provide the practical tools and scientific knowledge for studying the specificity and plasticity of human brain.

脑功能指纹在个体脑功能内在结构揭示、个体脑图谱绘制、脑科疾病极早期诊断等研究上具有重要科学价值。然而，当前基于功能磁共振成像的脑功能指纹识别存在脑活动解码不准确、特征区分性不高、分类器抗变异性不强及识别速度慢等问题。鉴于此，本项目在快速脑功能指纹识别理论、模型及算法上展开研究：首先，研究多功能属性融合的脑功能连通性检测模型及求解算法，通过对功能分离、功能整合、连接简约性、皮层稀疏编码以及皮层多响应性建模，来精确解码脑活动，以准确捕捉脑网络共性与个体特异性；其次，研究多域脑功能特征提取方法，获取在时间域、空间域及功能连接组等层面的脑功能特征，以全面刻画个体脑功能内在结构；再次，研究多域融合脑功能指纹选择及分类模型，自动挑选关键脑指纹，并进行精确识别；最后，研究脑功能指纹识别并行算法库，实现快速脑指纹识别。本项目所研究的脑功能指纹识别模型与方法，可为人脑特异性、可塑性等研究提供工具与知识积累。

脑功能指纹可分为群体水平脑功能指纹与个体水平脑功能指纹，对脑功能内在结构揭示、脑图谱绘制、脑科疾病极早期诊断等方面具有重要科学价值。本项目以脑功能指纹建模与识别、脑功能指纹应用为主线展开研究。就脑功能指纹建模与识别研究而言，针对群体水平脑功能指纹建模，提出了改进的动态脑功能连接组特征化模型、基于频谱差异映射框架的脑活动解码模型、基于傅里叶同步压缩变换脑低频波动的精细脑图谱划分模型以及基于人脑多特征图谱的脑活动状态识别模型等；针对个体水平脑功能指纹建模，提出基于脑特征图谱的多级联独立成分分析模型、多粒度脑功能图谱划分模型；以上研究有效解决了脑功能指纹微弱难于提取、特征区分性不高、识别模型鲁棒性不强等问题。就脑功能指纹应用研究而言，首先以人脑职业可塑性的内在神经生理学机制为研究点，采用脑功能指纹分析技术，以多种职业为研究对象，解析了人类脑职业可塑性的共性神经生理基础，同时也以职业船员为特定研究对象，提取了船员职业依赖的脑功能连接图谱，为职业神经科学发展提供了重要的知识积累；以中文书写的神经生理学机制为研究点，解析了支撑中文书写过程的共性脑功能连接指纹，同时也进一步厘清了中文书写过程中性别依赖的脑功能图谱，为人类书写的神经生理学机制解析提供了重要证据。进一步，研究团队将本项目所提出的脑功能指纹模型进行系统化，形成了多款软件工具包与系统，助力后续科研成果转化。最后，为抗击新冠疫情，促进复工复产，研究团队还开发了一款脑侦探心理评估系统，该系统面向群体为各大医院医生及精神心理科门诊患者，采用线上评估和智能分析技术，辅助医生有效地监控受试者心理健康状态，助力维护医院一线医师心理健康，同时避免防控诊治过程中接触性疫情传播，打造无接触“护心盾”，得到连云港市工信处的推广。