基于扩展的低阶多元广义线性模型的脑节点识别方法
|
摘要
针对现有单节点模型识别准确度较低以及低阶多元广义线性模型(LRMGLM)计算时间过长和使用局限性问题,提出基于扩展的低阶多元广义线性模型(ELRMGLM)的脑节点识别方法。首先,建立可以同时处理两次实验所有节点数据的ELRMGLM,以更多的时间空间信息来提高算法的准确度;然后,利用带时空平滑惩罚项的优化函数引入先验信息,并通过迭代函数对模型参数进行求解;最后,使用基于K-means的快速选择策略实现惩罚参数和大脑节点的快速选择。三次样本实验中,ELRMGLM的准确度分别比经典血液动力学响应函数(canonical)方法、平滑有限脉冲响应(SFIR)方法、正则化和广义交叉验证(Tik-GCV)方法的最优结果提升了约20%、8%、20%,略优于LRMGLM,且计算时间是LRMGLM的1/750。实验结果表明,ELRMGLM能有效提高大脑节点的识别准确度,减少计算时间。
Identifying brain nodes with different responses under different conditions plays an important role in human brain research. Due to the low detection accuracy of existing single-voxel models and the excessive calculation time and usage limitations of the Low-rank Multivariate General Linear Model( LRMGLM), a brain node identification method based on Extended LRMGLM( ELRMGLM) was proposed. Firstly, an ELRMGLM that can simultaneously process all node data in two experiments was established to improve the accuracy of the algorithm with more time and space information. Then, an optimization function with spatio-temporal smoothing penalty terms was used to introduce the prior information and the model parameters were solved through the iterative algorithm. Finally, a quick selection strategy based on K-means clustering was adopted to speed up penalty parameter selection and brain node identification. In three sample experiments, the accuracy of ELRMGLM was respectively increased by about 20%, 8% and 20% compared with that of canonical Hemodynamic Response Function( HRF) method( canonical), Smooth Finite Impulse Response( SFIR) and Tikhonov-regularization and GeneralizedCross-Validation( Tik-GCV), which was slightly better than LRMGLM. However, the calculation time of ELRMGLM was 1/750 of that of LRMGLM. The experimental results show that ELRMGLM can effectively improve the identification accuracy and reduce the calculation time.
Identifying brain nodes with different responses under different conditions plays an important role in human brain research. Due to the low detection accuracy of existing single-voxel models and the excessive calculation time and usage limitations of the Low-rank Multivariate General Linear Model( LRMGLM), a brain node identification method based on Extended LRMGLM( ELRMGLM) was proposed. Firstly, an ELRMGLM that can simultaneously process all node data in two experiments was established to improve the accuracy of the algorithm with more time and space information. Then, an optimization function with spatio-temporal smoothing penalty terms was used to introduce the prior information and the model parameters were solved through the iterative algorithm. Finally, a quick selection strategy based on K-means clustering was adopted to speed up penalty parameter selection and brain node identification. In three sample experiments, the accuracy of ELRMGLM was respectively increased by about 20%, 8% and 20% compared with that of canonical Hemodynamic Response Function( HRF) method( canonical), Smooth Finite Impulse Response( SFIR) and Tikhonov-regularization and GeneralizedCross-Validation( Tik-GCV), which was slightly better than LRMGLM. However, the calculation time of ELRMGLM was 1/750 of that of LRMGLM. The experimental results show that ELRMGLM can effectively improve the identification accuracy and reduce the calculation time.
引文
[1]FRISTON K J,HOLMES A P,WORSLEY K J,et al.Statistical parametric maps in functional imaging:a general linear approach[J].Human Brain Mapping,1994,2(4):189-210.
[2]FRISTON K,JEZZARD P,TURNER R.Analysis of functional MRItime-series[J].Human Brain Mapping,1994,1(2):15-171.
[3]WORSLEY K J,LIAO C H,ASTON J,et al.A general statistical analysis for f MRI data[J].Neuroimage,2002,15(1):1-15.
[4]GOUTTE C,NIELSEN F,HANSEN L.Modeling the hemodynamic response in f MRI using smooth FIR filters[J].IEEE Transactions on Medical Imaging,2000,19(12):1188-1201.
[5]CASANOVA R,YANG L,HAIRSTON W D,et al.Evaluating the impact of spatio-temporal smoothness constraints on the BOLD hemodynamic response function estimation:an analysis based on Tikhonov regularization[J].Physiological Measurement,2009,30(5):37-51.
[6]ZHANG T,LI F,BECKES L,et al.A semi-parametric model of the hemodynamic response for multi-subject f MRI data[J].Neuroimage,2013,75(4):136-145.
[7]ZHANG T,LI F,GONZALEZ M Z,et al.A semi-parametric nonlinear model for event-related f MRI[J].Neuroimage,2014,97(2):178-187.
[8]VINCENT T,RISSER L,CIUCIU P.Spatially adaptive mixture modeling for analysis of f MRI time series[J].IEEE Transanctions on Medical Imaging,2010,29(4):1059-1074.
[9]MAKNI S,CIUCIU P,IDIER J,et al.Joint detection-estimation of brain activity in functional MRI:a multichannel deconvolution solution[J].IEEE Transanctions on Signal Processing,2005,53(9):3488-3502.
[10]MAKNI S,IDIER J,VINCENT T,et al.A fully Bayesian approach to the parcel-based detection-estimation of brain activity in f MRI[J].Neuroimage,2008,41(3):941-969.
[11]CHAARI L,FORBES F,VINCENT T,et al.Hemodynamic-informed parcellation of f MRI data in a joint detection estimation framework[C]//Proceedings of the 2012 International Conference on Medical Image Computing and Computer-Assisted Intervention,LNCS 7512.Berlin:Springer,2012:180-188.
[12]DEGRAS D,LINDQUIST M A.A hierarchical model for simultaneous detection and estimation in multi-subject f MRI studies[J].Neuroimage,2014,98(7):61-72.
[13]ZHANG L,GUINDANI M,VERSACE F,et al.A spatio-temporal nonparametric Bayesian variable selection model of f MRI data for clustering correlated time courses[J].Neuroimage,2014,95(8):162-175.
[14]ZHANG L,GUINDANI M,VERSACE F,et al.A spatiotemporal nonparametric Bayesian model of multi-subject f MRI data[J].Annals of Applied Statistics,2016,10(2):638-666.
[15]ZHANG T,PHAM M,SUN J,et al.A low-rank multivariate general linear model for multi-subject f MRI data and a non-convex optimization algorithm for brain response comparison[J].Neuroimage,2017,173:580-591.
[16]ARBABSHIRANI M R,HAVLICEK M,KIEHL K A,et al.Functional network connectivity during rest and task conditions:a comparative study[J].Human Brain Mapping,2013,34(11):2959-2971.
[17]CALHOUN V D,ADALI T.Unmixing f MRI with independent component analysis[J].IEEE Engineering in Medicine and Biology Magazine,2006,25(2):79-90.
[18]WAHBA G.Spline Models for Observational Data,CBMS-NSFRegional Conference Series in Applied Mathematics[M].New York:Watson Research Center,1990:59.
[19]REISS P T,OGDEN R T.Functional principal component regression and functional partial least squares[EB/OL].[2018-02-03].http://wiki.sfu.ca/research/datagroup/images/1/1e/Paper.pdf.
[20]REISS P T,OGDEN R T.Smoothing parameter selection for a class of semiparametric linear models[J].Journal of the Royal Statistical Society,2009,71(2):505-523.
[21]WOOD S N.Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models[J].Journal of the Royal Statistical Society,2011,73(1):3-36.
[22]COAN J A.Adult attachment and the brain[J].Journal of Social and Personal Relationships,2010,27(2):210-217.
[23]COAN J.The Social Regulation of Emotion[M].New York:Oxford University Press,2011:614-623.
[24]COAN J A,SCHAEFER H S,DAVIDSON R J.Lending a hand:social regulation of the neural response to threat[J].Psychological Science,2006,17(12):1032-1039.
[2]FRISTON K,JEZZARD P,TURNER R.Analysis of functional MRItime-series[J].Human Brain Mapping,1994,1(2):15-171.
[3]WORSLEY K J,LIAO C H,ASTON J,et al.A general statistical analysis for f MRI data[J].Neuroimage,2002,15(1):1-15.
[4]GOUTTE C,NIELSEN F,HANSEN L.Modeling the hemodynamic response in f MRI using smooth FIR filters[J].IEEE Transactions on Medical Imaging,2000,19(12):1188-1201.
[5]CASANOVA R,YANG L,HAIRSTON W D,et al.Evaluating the impact of spatio-temporal smoothness constraints on the BOLD hemodynamic response function estimation:an analysis based on Tikhonov regularization[J].Physiological Measurement,2009,30(5):37-51.
[6]ZHANG T,LI F,BECKES L,et al.A semi-parametric model of the hemodynamic response for multi-subject f MRI data[J].Neuroimage,2013,75(4):136-145.
[7]ZHANG T,LI F,GONZALEZ M Z,et al.A semi-parametric nonlinear model for event-related f MRI[J].Neuroimage,2014,97(2):178-187.
[8]VINCENT T,RISSER L,CIUCIU P.Spatially adaptive mixture modeling for analysis of f MRI time series[J].IEEE Transanctions on Medical Imaging,2010,29(4):1059-1074.
[9]MAKNI S,CIUCIU P,IDIER J,et al.Joint detection-estimation of brain activity in functional MRI:a multichannel deconvolution solution[J].IEEE Transanctions on Signal Processing,2005,53(9):3488-3502.
[10]MAKNI S,IDIER J,VINCENT T,et al.A fully Bayesian approach to the parcel-based detection-estimation of brain activity in f MRI[J].Neuroimage,2008,41(3):941-969.
[11]CHAARI L,FORBES F,VINCENT T,et al.Hemodynamic-informed parcellation of f MRI data in a joint detection estimation framework[C]//Proceedings of the 2012 International Conference on Medical Image Computing and Computer-Assisted Intervention,LNCS 7512.Berlin:Springer,2012:180-188.
[12]DEGRAS D,LINDQUIST M A.A hierarchical model for simultaneous detection and estimation in multi-subject f MRI studies[J].Neuroimage,2014,98(7):61-72.
[13]ZHANG L,GUINDANI M,VERSACE F,et al.A spatio-temporal nonparametric Bayesian variable selection model of f MRI data for clustering correlated time courses[J].Neuroimage,2014,95(8):162-175.
[14]ZHANG L,GUINDANI M,VERSACE F,et al.A spatiotemporal nonparametric Bayesian model of multi-subject f MRI data[J].Annals of Applied Statistics,2016,10(2):638-666.
[15]ZHANG T,PHAM M,SUN J,et al.A low-rank multivariate general linear model for multi-subject f MRI data and a non-convex optimization algorithm for brain response comparison[J].Neuroimage,2017,173:580-591.
[16]ARBABSHIRANI M R,HAVLICEK M,KIEHL K A,et al.Functional network connectivity during rest and task conditions:a comparative study[J].Human Brain Mapping,2013,34(11):2959-2971.
[17]CALHOUN V D,ADALI T.Unmixing f MRI with independent component analysis[J].IEEE Engineering in Medicine and Biology Magazine,2006,25(2):79-90.
[18]WAHBA G.Spline Models for Observational Data,CBMS-NSFRegional Conference Series in Applied Mathematics[M].New York:Watson Research Center,1990:59.
[19]REISS P T,OGDEN R T.Functional principal component regression and functional partial least squares[EB/OL].[2018-02-03].http://wiki.sfu.ca/research/datagroup/images/1/1e/Paper.pdf.
[20]REISS P T,OGDEN R T.Smoothing parameter selection for a class of semiparametric linear models[J].Journal of the Royal Statistical Society,2009,71(2):505-523.
[21]WOOD S N.Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models[J].Journal of the Royal Statistical Society,2011,73(1):3-36.
[22]COAN J A.Adult attachment and the brain[J].Journal of Social and Personal Relationships,2010,27(2):210-217.
[23]COAN J.The Social Regulation of Emotion[M].New York:Oxford University Press,2011:614-623.
[24]COAN J A,SCHAEFER H S,DAVIDSON R J.Lending a hand:social regulation of the neural response to threat[J].Psychological Science,2006,17(12):1032-1039.