Linear parameter-varying model and adaptive filtering technique for detecting neuronal activities: an fNIRS study
Linear parameter-varying model and adaptive filtering technique for detecting neuronal activities: an fNIRS study
复制标题
DOI：
10.1088/1741-2560/10/5/056002
复制
发表时间：
2013-10-01
影响因子：
4
通讯作者：
Hong, Keum-Shik
中科院分区：
文献类型：
--
作者：
Kamran, M. Ahmad;Hong, Keum-Shik
Objective.功能性近红外光谱（fNIRS）是一种新兴的非侵入性脑成像技术，其通过使用650-950 nm波长的近红外光来测量脑活动。fNIRS的主要优点是它的低成本，便携性，和良好的时间分辨率作为一个合理的解决方案，实时成像。最近的研究表明，fNIRS作为脑机接口工具的巨大潜力。Approach.本文提出了第一个新的技术，fNIRS为基础的建模的大脑活动，使用线性参数变化（LPV）的方法和自适应信号处理。每个通道的输出信号被假定为具有由仿射投影算法最优估计的未知系数的LPV系统的输出。假设参数向量是高斯的。主要结果。广义线性模型（GLM）是一种常用的功能磁共振成像数据分析方法，但它在光信号的情况下有一定的局限性。所提出的模型是更有效的意义上说，它允许用户定义更多的状态。此外，与大多数以前的模型不同，它是在线的。目前的研究结果，显示出改善，在广泛的实验中通过随机手指敲击任务进行了验证。我们使用了24个状态，可以根据计算成本和要求减少或增加。意义采用t统计量确定激活图并验证结果的显著性。所提出的技术和现有的两个基于GLM的算法的比较表明，在估计血流动力学响应的改进。此外，所提出的算法的收敛性表明，在连续迭代的误差减少。
Objective. Functional near-infrared spectroscopy (fNIRS) is an emerging non-invasive brain imaging technique that measures brain activities by using near-infrared light of 650-950 nm wavelength. The major advantages of fNIRS are its low cost, portability, and good temporal resolution as a plausible solution to real-time imaging. Recent research has shown the great potential of fNIRS as a tool for brain-computer interfaces. Approach. This paper presents the first novel technique for fNIRS-based modelling of brain activities using the linear parameter-varying (LPV) method and adaptive signal processing. The output signal of each channel is assumed to be an output of an LPV system with unknown coefficients that are optimally estimated by the affine projection algorithm. The parameter vector is assumed to be Gaussian. Main results. The general linear model (GLM) is very popular and is a commonly used method for the analysis of functional MRI data, but it has certain limitations in the case of optical signals. The proposed model is more efficient in the sense that it allows the user to define more states. Moreover, unlike most previous models, it is online. The present results, showing improvement, were verified by random finger-tapping tasks in extensive experiments. We used 24 states, which can be reduced or increased depending on the cost of computation and requirements. Significance. The t-statistics were employed to determine the activation maps and to verify the significance of the results. Comparison of the proposed technique and two existing GLM-based algorithms shows an improvement in the estimation of haemodynamic response. Additionally, the convergence of the proposed algorithm is shown by error reduction in consecutive iterations.