Klinische Neurophysiologie 2010; 41 - ID78
DOI: 10.1055/s-0030-1250907

On the Estimation of the Direction of Information Flow

B Schelter 1, L Sommerlade 1, F Amtage 2, O Lapp 1, B Hellwig 2, CH Luecking 2, J Timmer 1, 3
  • 1University of Freiburg, Center for Data Analysis and Mod., Freiburg, Deutschland
  • 2University Medical Center, Department of Neurology, Freiburg, Deutschland
  • 3University of Freiburg, Freiburg Institute for Advanced Studies, Freiburg, Deutschland

Introduction: The analysis of signals in the neurosciences raises several questions. One important issue is the detection of interactions between signals, as they promise to disclose the biological basis underlying the normal behavior or the abnormal functioning of certain networks. Here, it is of particular importance to reliably infer the direction of information flow. Some approaches to this aim have been suggested. We compare their abilities and limitations on simulated data as well as in an application to essential tremor.

Material and Methods: Based on simulated data we investigate the abilities and limitations of frequently used analysis techniques which were suggested being capable of detecting the direction of information flow. The techniques are the maximum coherence approach (Govindan, R. B., et al., 2005. Estimation of time delay by coherence analysis. Physica A 350, 277–295), an approach based on the Hilbert phase (Timmermann, L., et al., 2003. The cerebral oscillatory network of Parkinsonian resting tremors. Brain 126, 199–212), the partial directed coherence (Baccala, L. A., Sameshima, K., 2001. Partial directed coherence: a new concept in neural structure determination. Biol. Cybern. 84, 463–474), and the partial directed correlation (Eichler, M., 2006. Graphical modeling of dynamic relationships in multivariate time series. In: Schelter, B., Winterhalder, M., Timmer, J. (Eds.), Handbook of Time Series Analysis. Wiley-VCH, Ch. 14, pp. 335–372). In an application to six patients suffering from essential tremor we show the applicability to actual data sets.

Results: By means of a simulation study we characterize the abilities and limitations of the individual approaches. It turned out that the directed partial correlation approach is the most powerful one, allowing both to estimate the direction of information flow as well as the time lag reliably. In the application to essential tremor we could clearly demonstrate that tremor correlated cortical activity is composed of both directions, i.e. from the cortex to the muscles and vice versa.

Conclusion: Although several techniques have been suggested in the literature addressing the same aim, i.e. to estimate the direction of information flow, they perform considerably different both on simulated as well as on actual data. Directed partial correlation turned out to be the best performing one.