Cortés, J. 1, 2 , Erramuzpe, A. 1, 2 , Escudero, I. 1, 3 , Mateos, B. . 1, 3 , Cabrera, A. . 4 , Marinazzo, D. 5 , Sanz-Arigita, E. J. 6 , Stramaglia, S. . 1, 2, 7, 8 & Cortes, J. M. . 1, 2, 8
1 Computational Neuroimaging Lab. Biocruces Health Research Institute. Cruces University Hospital, Barakaldo, Spain
2 Quantitative Biomedicine Unit. Biocruces Health Research Institute. Cruces University Hospital, Barakaldo, Spain
3 Radiology Service. Cruces University Hospital, Barakaldo, Spain
4 Osatek. Bilbao, Spain
5 Faculty of Psychology and Educational Sciences, Department of Data Analysis, Ghent University, Belgium
6 Department of Radiology, CITA-Alzheimer Foundation, San Sebastian, Spain
7 Dipartimento di Fisica, Universit degli Studi di Bari and INFN, Bari, Italy
8 Ikerbasque, The Basque Foundation for Science. Bilbao, Spain
The resting brain dynamics self-organizes in a finite number of correlated patterns known as resting state networks (RSNs). Currently it is well-known that techniques like independent component analysis can separate the brain?s resting state functional magnetic resonance imaging activity to provide such RSNs but the specific pattern of interaction between these RSNs is not well understood yet. To this aim, we propose here a novel method to compute the information flow between different RSNs. Our method first uses principal component analysis to reduce dimensionality in each RSN to then compute the information flow between the different RSNs by systematically increasing k (the number of principal components used in the calculation). When k=1, this method is equivalent to computing the information flow between the average voxels activity in each RSN. For k>= 1, our method calculates the k-multivariate transfer entropy between the different RSNs. Our results are two-fold: first, the pattern of information flow was dimension dependent, increasing from k=1 (average voxels activity) up to reach a maximum at k=5 to finally decay to zero for k bigger than 10. Therefore, the total amount of transferred information does not grow monotonically with the dimension. Second, we have applied our method to compare inter-RSNs information transfer between Alzheimer patients and controls. The more significant differences in between AD and control groups occurred for k=2. This allowed us to perform a paired t-test to spatially localize the k=2 component, thereby identifying within-RSN brain regions related to differential information transfer between Alzheimer?s and controls. We conclude that this methodology can unravel aspects regarding the resting state organization and dynamics of the Alzheimer disease.
Work supported by Ikerbasque: The Basque Foundation for Science, Gobierno Vasco (Saiotek SAIO13-PE13BF001) and Euskampus at UPV/EHU.