ORIGINAL PAPER Parcel-Based Connectivity Analysis of fMRI Data for the Study of Epileptic Seizure Propagation Maria Gabriella Tana • Anna Maria Bianchi • Roberta Sclocco • Tiziana Franchin • Sergio Cerutti • Alberto Leal Received: 29 November 2011 / Accepted: 14 March 2012 / Published online: 5 April 2012 Ó Springer Science+Business Media, LLC 2012 Abstract The aim of this work is to improve fMRI Granger Causality Analysis (GCA) by proposing and comparing two strategies for defining the topology of the networks among which cerebral connectivity is measured and to apply fMRI GCA for studying epileptic seizure propagation. The first proposed method is based on infor- mation derived from anatomical atlas only; the other one is based on functional information and employs an algorithm of hierarchical clustering applied to fMRI data directly. Both methods were applied to signals recorded during seizures on a group of epileptic subjects and two connec- tivity matrices were obtained for each patient. The per- formances of the different parcellation strategies were evaluated in terms of their capability to recover informa- tion about the source and the sink of the network (i.e., the starting and the ending point of the seizure propagation). The first method allows to clearly identify the seizure onset in all patients, whereas the network sources are not so immediately recognizable when the second method was used. Nevertheless, results obtained using functional clus- tering do not contradict those obtained with the anatomical atlas and are able to individuate the main pattern of propagation. In conclusion, the way nodes are defined can influence the easiness of identification of the epileptogenic focus but does not produce contradictory results showing the effectiveness of proposed approach to formulate hypothesis about seizure propagation at least in the early phase of investigation. Keywords Functional MRI  Parcellation  Brain networks  Connectivity  Granger causality Introduction Functional Magnetic Resonance Imaging (fMRI) is increasingly used to investigate the functional interactions within networks of brain areas that support specific cog- nitive or sensorimotor task or that are involved in patho- logical processes (Marrelec et al. 2006). The problem of investigating inter-region connectivity can be approached through two main classes of analysis techniques: model-based methods (e.g., Dynamical Causal Modeling or DCM) (Friston et al. 2003) and data-driven methods (e.g., Granger Causality Analysis or GCA) (Roebroeck et al. 2011). GCA has been increasingly used in neuroimaging since it does not require any pre-specification or a priori knowledge about the connectivity structure and it has been successfully applied to fMRI data for studying brain M. G. Tana (&)  A. M. Bianchi  R. Sclocco  T. Franchin  S. Cerutti Department of Bioengineering, Politecnico di Milano, Via C.Golgi 39, 20133 Milan, Italy e-mail: mariagabriella.tana@polimi.it Present Address: M. G. Tana BIND - Behavioral Imaging and Neural Dynamics Center, University ‘‘G. d’Annunzio’’ of Chieti-Pescara, Via dei Vestini 33, 66100 Chieti (CH), Italy T. Franchin Bambino Gesu ` Children Hospital, IRCCS, P.zza S.Onofrio 4, 00165 Rome, Italy A. Leal Department of Neurophysiology, Centro Hospitalar Psiquia ´trico de Lisboa, Av. do Brasil 53, 1749-002 Lisbon, Portugal A. Leal Centre for Psychological Research and Social Intervention (CIS-IUL), Av. das Forc ¸as Armadas, 1649-026 Lisbon, Portugal 123 Brain Topogr (2012) 25:345–361 DOI 10.1007/s10548-012-0225-2