人腦功能性核磁共振影像之有效性連結分析

Effective connectivity analysis on functional magnetic resonance imaging of the human brain

朱盈樺(Ying-Hua Chu)

指導教授 ： 林發暄

國立臺灣大學/工學院/醫學工程學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.03448

摘要

本研究在於發展功能性核磁共振影像 (functional magnetic resonance imaging, fMRI) 之有效性連結分析方法，以闡明人腦中訊息傳遞的因果關係。分析方法從參數模型的Granger causality (GCAR)至無模型的消息理論 (information theory)，消息理論包含time delayed mutual information (MI)和transfer entropy (TE)。這些方法可以從雙變數推展至多變數分析，配合統計方法，提高辨別訊息傳遞方向的特異性 (specificity)。使用amplitude adjusted Fourier-transformed (AAFT) 演算方法，製造出統計所需的虛無分配 (null distribution)，再進行消息方向可信度的虛無假設。在本研究的模擬針對雙變數的分析，雖然可以正確辨別方向，但在多變數時間序列的環境下，無論是GCAR、time delayed MI或TE，單單使用雙變數方法分析應無法避免間接傳遞的訊息，因此，配合適當的條件設定可避免上述情況發生。但綜合模擬結果，TE配合AAFT可以提供最好的特異性 (或最低的type I error)。這些分析方法也應用在真實的fMRI 實驗上，運用ultra-fast magnetic inverse imaging (InI) 方法取得高取樣的時間序列 (10 Hz)，透過time delayed MI得到各活動腦區的時間延遲量。TE則估計出消息傳遞的方向。因此，於fMRI認知實驗上，配合使用time delayed MI和TE，將可以有效得到人腦活動區域中的因果關係。

關鍵字

Granger causality ； transfer entropy ； functional magnetic resonance imaging (fMRI) ； effective connectivity

並列摘要

The purpose of this thesis is to develop data-driven effective connectivity analysis tools to reveal causal relationships in the human brain using functional magnetic resonance imaging (fMRI) measurements. I study Granger causality (GCAR) and propose the information theory-based methods, including time delayed mutual information (MI) and transfer entropy (TE). These methods can process fMRI data using either a bivariate or a multivariate approach to respectively obtain efficient calculation or to improve the specificity of the causality detection. Also, to provide statistical inference, I propose to empirically estimate the null distribution of causality measures by the amplitude adjusted Fourier-transformed (AAFT) algorithm.
Provided with two coupled time series, My simulations show that the pair-wise GCAR, time delayed MI, and TE can distinguish the information flow, but can not avoid false causality estimation due to the potential indirect information flow. Therefore, appropriate conditioning is crucial to control the specificity of causality estimation. Simulation on the fMRI time series model suggests that compared to GCAR, TE combined with AAFT has a higher specificity (a lower type I error rate).
I also apply effective connectivity analysis to in vivo visuomotor fMRI experiments using ultra-fast magnetic inverse imaging (InI). Facilitated with a high volumetric sampling rate of 10 Hz, time delayed MI found the latency between activated brain areas in the lateralized conditions. TE further estimates potential bi-directional and uni-direction causal modulation in the lateralized visuomotor network. I conclude that the time delayed MI and TE can be useful tools to delineate causal interactions in spatiotemporal imaging of human brain during tasks and cognition.

並列關鍵字

Granger causality ； transfer entropy ； functional magnetic resonance imaging (fMRI) ； effective connectivity

參考文獻

2. McIntosh, A.R. and F. Gonzalez-Lima, Structural equation modeling and its application to network analysis in functional brain imaging. Human Brain Mapping, 1994. 2(1-2): p. 2-22.

3. Bullmore, E., et al., How good is good enough in path analysis of fMRI data? Neuroimage, 2000. 11(4): p. 289-301.

4. Friston, K.J., L. Harrison, and W. Penny, Dynamic causal modelling. Neuroimage, 2003. 19(4): p. 1273-1302.

5. Roebroeck, A., E. Formisano, and R. Goebel, The identification of interacting networks in the brain using fMRI: Model selection, causality and deconvolution. Neuroimage, 2009. In Press, Corrected Proof.

6. Granger, C.W.J., Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 1969. 37(3): p. 414-&.