作为一种内源性神经调控技术，神经反馈利用脑成像技术采集用户的脑活动，并通过图形、声音等直观形式实时反馈给用户，使他们在多次训练后可以找到恰当的策略实现对脑活动的自主调控，并带来脑可塑性及相应行为的改变。基于近红外成像的神经反馈具有高生态效度、低限制性、便携易用、兼具时空间分辨率等特点，在长期日常训练及临床治疗中均表现出巨大潜力。 然而，目前的近红外神经反馈（fNIRS-NF）研究存在三方面亟待优化的问题：1）头壳定位精度优化问题。近红外技术通过在被试头壳上放置光极来探测其覆盖区域的局部脑活动。考虑到成像局部性，能否根据想要调节的脑区（靶脑区）在被试头壳上精确定位出光极的放置位置直接决定了fNIRS-NF能否准确地调节靶脑区活动。此外，在多次训练中保持头壳靶点位置的一致性也对神经反馈的效果至关重要。靶点的精确性和一致性都依赖于进一步提高头壳定位精度，以使用户能准确地描述、定位并导航至头壳上任意点。2）反馈形式优化问题。fNIRS-NF需要进行多次训练，而现有的温度计等反馈形式单调无趣，容易导致用户依从性差，主动性低，儿童群体难以集中注意力，从而影响反馈效果甚至使训练无法继续。3）软件平台优化问题。在fNIRS-NF训练过程中，软件平台承担了操纵并监控实验流程，与近红外设备实时通讯获取信号并进行实时滤波、反馈指标计算等任务。早期的软件平台基于MATLAB开发，绘图速度慢且难以多线程运行，难以开发交互界面，因此无法满足神经反馈在实时性和简便性上的需求。 针对上述问题，本论文的主要研究内容如下： 1）系统分析了头壳定位的误差来源并对其进行优化。我们将头壳定位误差归因至两类误差源：一类是在三维定位仪使用过程中由于操作不当产生的误差，对此本文评估了定位仪使用中各环节的操作误差并确定了使误差最小化的操作规范；第二类是头壳重建误差。头壳重建由四个头壳基准点及多个头壳表面离散点经三维表面重建技术完成。基准点及离散点采集的准确性是影响头壳重建误差的关键。基准点中，枕骨隆凸点的生理定义不清晰，手动量取常产生偏差。本文模拟评估了枕骨隆凸偏差对头壳重建的影响，并创新性地提出了枕骨隆凸替代点自动量取算法。离散采点过程中，采点的数量、均匀性、野点都会影响头壳重建结果。本文给出了采点数量恰当、均匀的优化采点方案。为了进一步帮助研究者量化评估头壳重建结果并对离群点进行预警，保障头壳重建稳定性，我们提出了基于PCA分解的群体头壳形状分析算法。最后，本文对比了优化前后三名被试在五次实验中的重复头壳定位误差，验证了优化方法的效果。 2）引入游戏（意念飞剑）、实物（机器人）、社会交互（双人合作与竞争）等因素，提升了反馈形式的多样性与趣味性。并在实际实验中测试了优化效果。 3）设计基于python多线程框架的fNIRS-NF软件平台，其中有包含多类fNIRS设备的通讯兼容、允许用户全面自定义各种实验参数、提供快速交互的监控界面等优点。该平台运行速度快、交互便捷，满足研究者的使用需求。 最后，我们以注意缺陷多动障碍儿童的神经反馈研究为例，介绍本文优化的近红外神经反馈系统的实际应用。实验结果表明，经过多次训练后患者的认知功能有较大提升。从而验证了优化后神经反馈系统的可行性和有效性。

As an intrinsic Neuromodulation techonology, Neurofeedback gets user’s brain activityis by Neuroimaging and show them back to user real time in picture or voice modal. So they can find properly strategies to regulate their own brain activity, which can change brain plasticity and behaviors. While Neurofeedback based on functional Near Infrared Spectrum Instrument (fNIRS-NF) shows big potential in daily trainning and medical therapy, due to its high ecological validity , less limitation, portability and good time-space resolution. But fNIRS-NF has three problems that need to be optimized urgently. First,we should improve the accuracy of scalp positioning. fNIRS detects activity of local brain area which is under the optodes located in subject’s scalp. Taking fator of locality, to get precisely position for the placement of optodes on subjects’ scalp on the basis of region of interests(Target Brain Area) , determines how fNIRS-NF can regulates activity of Target Barin Area accurately. Besides, maintaining the same position in the multiple NF trainning is also make a great difference to Neurofeedback. The accuracy and consistency above depend on better scalp positioning precision, to make user describe,locate and navigate to any point on scalp. Second, we should improve the Neurofeedback modal. Subjects need to work continuely in NF trainning. Existing boring modal(like thermometer) leads to less obedience and motivation of subjects. Especially Group of children focus on so difficultly that Neurofeedback has poor effect or even can’t continue. Third, some software or toolbox undertakes tasks of operating and monitoring experiment. It should be used to communicate with fNIRS Mechine real-time, to filrt and compute real-time signal. But older version software-platform developed on MATLAB is hard to run multi-thread and develop GUI, which leads to slow speed and complex operation. It can’t fit needs of researchers. According to the three problems above, there is a main work of this thesis as follow. First, we analyse the sources of scalp positioning error and optimize them. We attribute error to two aspects. One is error 3D digitizer when we use it by mistakes. We access the error about it and lay down a proper operation specification. The other source is scalp reconstruction, which is finished by 4 scalp references, disperse scatters and 3d surface reconstruction algorithm. The accuracy of references and scatters is the key of scalp reconstruction. As a reference, the Occipital Protuberance(Iz) is not well defined physiologically, which always leads to siting bias. In this thesis we assess the scalp drift influence by Iz shift and raise an algorithm to get the position of Iz automaticly. For scatters, number of points, homogeneity, outilers influence. We present a suitable siting solution with right number and uniformity. To help researchers to evaluate the result of scalp reconstruction quantifily and get hints of the outliers early , We present a scalp shape analysis algorithm based on PCA decomposition, which can guarantee stability of scalp reconstrution.Finally, we compared the error between a pre-optimized appproach and a post-optimized appproach, by 3 subjects and 5 repeated scalp positioning. The results prove effect of optimization. Second, we improve the diversity and enjoyment of Neurofeedback modal, in three aspects of gaming(TigerKiller), physical feedback(Robots) and social factors of two-people NF. And we test the effect of these feedback modal. Three, the fNIRS-NF software platform is designed and implemented, which is based on python multithreading framework, meeting the requirements of fast runing, esily interacting and conveniently operating. It includes features of multi-mechine compatibility, comprehensively setting and fast-interactive GUI. Finally, we introduces a Neurofeedback research of ADHD that used the optimized fNRIS-NF system in this thesis. It is indicated that subject improves cognitive function a lot after long time trainning. It proved the feasibility and effectivity of this system meaningly.