Detecting information flow direction in multivariate linear and nonlinear models

Chunfeng Yang (1) , Régine Le Bouquin-Jeannès (1) , Gérard Faucon (1) , Huazhong Shu (2, 3)

1 LTSI - Laboratoire Traitement du Signal et de l'Image
2 CRIBS - Centre de Recherche en Information Biomédicale sino-français
3 LIST - Laboratory of Image Science and Technology [Nanjing]

In this paper we present an approach to analyze the direction of information flow between time series involving bidirectional relations. The intuitive idea comes from a first study dedicated to the so-called phase slope index, which is a measure originally developed to detect unidirectional relations and is based on the complex coherence function. In order to detect bidirectional flows, we propose two new causality indices supplying the previous index with two other functions, the directed coherence function and the directed transfer function. Moreover, to cope with the inability of the approaches based on coherence (ordinary or directed) or on directed transfer function to distinguish between direct and indirect relations, we propose another causality index based on the partial directed coherence to identify only direct relations. Experimental results show that some challenges have promising solutions through the use of this new indicator dealing with both linear and nonlinear multivariate models.

Mots clés: Effective connectivity; Causal relations; Partial directed coherence; Information flow; Multivariate time series

Domaines: Ingénierie biomédicale; Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Chunfeng Yang, Régine Le Bouquin-Jeannès, Gérard Faucon, Huazhong Shu. Detecting information flow direction in multivariate linear and nonlinear models. Signal Processing, 2013, 93 (1), pp.304-312. ⟨10.1016/j.sigpro.2012.05.018⟩. ⟨inserm-00759802⟩