Journal of Integrative Neuroscience, Vol. 12, No. 4 (2013) 441–447 c Imperial College Press ° DOI: 10.1142/S021963521350026X
Brain network analysis of EEG functional connectivity during imagery hand movements
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
Matteo Demuru, Francesca Fara and Matteo Fraschini* Dipartimento di Ingegneria Elettrica ed Elettronica Universit a di Cagliari, Italia *[email protected] [Received 18 July 2013; Accepted 9 September 2013; Published 25 October 2013] The characterization of human neural activity during imaginary movement tasks represent an important challenge in order to develop e®ective applications that allow the control of a machine. Yet methods based on brain network analysis of functional connectivity have been scarcely investigated. As a result we use graph theoretic methods to investigate the functional connectivity and brain network measures in order to characterize imagery hand movements in a set of healthy subjects. The results of the present study show that functional connectivity analysis and minimum spanning tree (MST) parameters allow to successfully discriminate between imagery hand movements (both right and left) and resting state conditions. In conclusion, this paper shows that brain network analysis of EEG functional connectivity could represent an e±cient alternative to more classical local activation based approaches. Furthermore, it also suggests the shift toward methods based on the characterization of a limited set of fundamental functional connections that disclose salient network topological features. Keywords: Imagery movements; functional connectivity; brain networks; minimum spanning tree; EEG; beta band.
1. Introduction The characterization and description of human neural activity during imaginary movement tasks still represent an important challenge. Currently, great interest about both the neuro-rehabilitation techniques based on motor imagery (Mulder, 2007; Holper et al., 2010) and brain computer interface (BCI) systems based on the interpretation of brain activity, attempt to provide (partially or completely) paralyzed patients with a potentially new communication tool. Some very recent reviews (Makeig et al., 2012; Nicolas-Alonso & Gomez-Gil, 2012; Tangermann et al., 2012) outlined the state-of-the-art of BCI systems reporting and discussing the commonly used methods for the characterization of the recorded neurological activity. Methods based on brain network analysis of functional connectivity have been scarcely investigated (Brunner et al., 2006; Daly et al., 2012). These graph theoretical methods have been widely adopted to successfully describe human brain dynamics (Bullmore & Sporns, 2009; *Corresponding
author.
441
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
442
M. DEMURU, F. FARA & M. FRASCHINI
Stam & van Straaten, 2012). Although brain network approach represents an important tool that could be used to develop more e±cient BCI systems, it may not be su±cient to thoroughly describe human brain dynamics as just information °ow in a complex anatomical and functional network (Poznanski, 2009). The aim of the present paper is to investigate the potentiality of functional connectivity analysis and brain network measures in order to characterize imagery (right and left) hand movements in a set of healthy subjects. In particular, we aim to understand if this approach could successfully allow to discriminate between imagery right and left hand movements, and between imagery movements (both right and left) and resting state condition. Functional connectivity was estimated by using a well-known measure of phase synchronization, named phase lag index (PLI) (Stam et al., 2007), which also allowed controlling the relevant problem of volume conduction. Typical normalized clustering coe±cient and shortest path length were evaluated in order to describe the topology of the brain networks, and the use of the minimum spanning tree (MST) (Boersma et al., 2013) has been proposed as a characteristic functional backbone of the network and as a possible solution to unbiased network comparison. 2. Materials and Methods The dataset used in this paper, public and freely downloadable from the web (http:// physionet.org/pn4/eegmmidb/), was created and contributed by the developers of the BCI2000 instrumentation system (Goldberger et al., 2000; Schalk et al., 2004). The dataset consists of 64 channels EEG recordings obtained from 109 volunteers and is provided in EDFþ format. Subjects performed di®erent motor/imagery tasks while 64 channels EEG signals were recorded at 160 Hz sampling rate. Each subject performed 14 experimental runs: two one-minute baseline runs (one with eyes open, one with eyes closed), and three two-minute runs of each of the four following tasks: (i) the subject opens and closes ¯st; (ii) the subject imagines opening and closing the ¯st; (iii) the subject opens and closes either both ¯sts or both feet; (iv) the subject imagines opening and closing either both ¯sts or both feet. Each of the four described task was organized in multiple trials of 4.1 s alternating same length resting state trials. In our analysis we studied the task (ii), so taking into account only imagined right and left hand movement trials and resting state trials between the tasks. Eighteen trials (4,1 s long) were analyzed for each subject and condition (right hand, left hand, resting condition before right hand and resting condition before left hand). The signal analysis was based on estimating the phase synchronization of all pairs of sensors during each trial for each of the 105 subjects (four subjects were successively excluded from analysis because recording parameters were di®erent from the other subjects). The data was averaged over all trials of the same task and averaged over each repetition.
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
BRAIN NETWORK ANALYSIS OF EEG FUNCTIONAL CONNECTIVITY
443
The described analysis was performed for each of the following frequency bands: delta (0.5–3.5 Hz), theta (4–8 Hz), alpha (8–13 Hz) and beta (13–30 Hz). To evaluate the pair-wise phase coupling between EEG signals we estimated the corresponding PLI (Stam et al., 2007). PLI re°ects the consistency with which one signal is phase leading or lagging with respect to another signal, allowing evaluating the phase di®erences between two signals with a measure not a®ected by the in°uence of volume conduction. Besides a global approach (mean PLI calculation) a more regional approach was also successively used. EEG channels were grouped into each hemisphere, and average PLI for all channels within a hemisphere (right and left) were computed. From synchronization matrices, we analyzed fully weighted and undirected graphs by means of the clustering coe±cient (which denotes the likelihood that neighbors of a vertex will also be connected to each other) and the characteristic path length (i.e., the average number of edges of the shortest path between pairs of vertices). All metrics were successively normalized by using randomly rewired versions of the original networks (averaged over 50 repetitions) matched for degree distribution. Furthermore, from every connectivity matrix, the MST was computed (by Kruskal's algorithm) and from these MST graphs several measures were successively estimated: diameter (largest distance between any two nodes), eccentricity (longest distance between a node and any other node), leaf number (number of nodes with degree ¼ 1) and hierarchy (refers to balance in hub nodes). As already reported (Boersma et al., 2013), these measures allow describing and characterizing the topology of the related MST graphs. All the analysis was performed using the Brainwave software (version 0.9.70, http://home.kpn.nl/stam7883/brainwave.html). Statistical analysis was performed with PRISM 5 for Mac OS X version 5.0f (GraphPad Software, Inc.). Two-way ANOVA for repeated measures, using condition (imagery movement/resting state) and side (left/right hand) as within-subject factors, was used to test di®erences for each of the computed measures. Two-way ANOVA for repeated measures, using region (right/left hemisphere) and side (left/ right hand) as within-subject factors, was used to test di®erences for regional (hemispheric) synchronization. A signi¯cance level of 0.05 was used for the test. Successively, Sidak's multiple comparisons test was used to correct for multiple comparisons. 3. Results Global synchronization analysis (mean PLI over all channels) showed a signi¯cant e®ect of condition (imagery movement/resting state) (F½1; 104 ¼ 23:10, p < 0:0001) but no e®ect of side (left/right hand) in the beta band. Sidak's multiple comparisons test showed a signi¯cant global synchronization increase both for right (tð104Þ ¼ 5:344, p < 0:0001) and left (tð104Þ ¼ 3:755, p < 0:0001) imagery hand movement when compared with respective resting state conditions in the beta band (Table 1).
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
444
M. DEMURU, F. FARA & M. FRASCHINI
Regional synchronization analysis (mean PLI over all channels within hemisphere) showed a signi¯cant side per region interaction e®ect (F½1; 104 ¼ 20:47, p < 0:0001) and a signi¯cant regional e®ect (F½1; 104 ¼ 144:0, p < 0:0001) in the beta band. Sidak's multiple comparisons test showed a signi¯cant regional synchronization di®erence in the left hemisphere between right and left hand imagery movement (tð104Þ ¼ 4:398, p < 0:0001) in the beta band (Table 2). No variations for both global and regional synchronization analysis were found in the other frequency bands. No signi¯cant e®ects of condition (imagery movement/resting state) and side (left/right hand) were observed for both normalized clustering coe±cient and characteristic path length in any frequency band. Signi¯cant e®ect of condition (imagery movement/resting state) and side (left/ right hand) were observed for MST eccentricity (F½1; 104 ¼ 41:58, p < 0:0001 and F½1; 104 ¼ 7:345, p ¼ 0:0079, respectively) and MST diameter (F½1; 104 ¼ 40:92, p < 0:0001 and F½1; 104 ¼ 8:148, p ¼ 0:0052, respectively) in the beta band (see Fig. 1). Signi¯cant e®ect of condition (imagery movement/resting state) was observed for MST hierarchy (F½1; 104 ¼ 4:793, p ¼ 0:31) in the beta band. No significant e®ects of condition (imagery movement/resting state) and side (left/right hand) were observed for MST leaf in any frequency band.
Table 1. Results of global PLI in beta band. Statistics refers to post test. IM refers to imagery hand movement condition, RS refers to resting state condition. Left hd refers to left hand and Right hd refers to right hand. Condition
Side
Left hd Right hd
IM
RS
Statisitics
0.185 0.028 0.186 0.028
0.182 0.027 0.182 0.027
*** ****
***p > 0:0001. ****p < 0:00001.
Table 2. Results of regional PLI in beta band. Statistics refers to post test (NS for not signi¯cant). Left hm refers to left hemisphere region, Right hm refers to right hemisphere region. Left hd refers to left hand and Right hd refers to right hand. Side
Region
Left hm Right hm
***p < 0:0001.
Left hd
Right hd
Statistics
0.218 0.038 0.194 0.030
0.222 0.036 0.192 0.029
*** NS
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
BRAIN NETWORK ANALYSIS OF EEG FUNCTIONAL CONNECTIVITY
445
Fig. 1. Signi¯cant repeated measures ANOVA e®ects for MST parameters in beta band. Left side shows MST diameter values for condition and imagery movement hand side. Right side shows MST eccentricity values for condition and imagery movement hand side. Values are expressed as mean and standard deviation.
No di®erences between condition and side for any other network measures were observed. 4. Discussion The results of the present study show that functional connectivity analysis should be considered as an e±cient tool in order to characterize brain dynamics involved during imagery hand movement. Indeed, it has been successfully reported that phase synchronization analysis optimally allowed to discriminate between imagery hand movements (both right and left) and resting state conditions. Moreover, it seems that the left hemisphere regional synchronization mechanisms could play a more relevant role to discriminate between right and left hand movements. This ¯nding, which is not unexpected (Janssen et al., 2011), represents, in our opinion, probably the main result of this study and potentially also suggests a preferential area to be investigated during BCI and neuro-rehabilitation applications. An important issue that should be always addressed when dealing with EEG systems is related to volume conduction problem. Even if our approach (PLI) is particularly conservative, since it discards also possible real zero-lag synchronization, our results still show that synchronization mechanism can be used to discriminate EEG brain dynamics. Furthermore, brain network analysis, in particular the MST parameters characterization, showed that it is possible to observe network topology modi¯cations induced by imagery hand movements compared to resting state conditions. For the ¯rst time, to the best of our knowledge, in the present paper the problem to characterize brain dynamics related to imagery hand movement using a large set of brain network measures (Roy, 2012) and also (using the MST graphs) the overlooked relevant problem of network comparison has been addressed (Van Wijk et al., 2010). Our results strongly suggest that these parameters could be used to discriminate imagery movements from resting state conditions and that the MST could identify a functional backbone to be preferentially investigated in any application. It also should not surprise that our results were observed in the beta (13–30 Hz) band, since
446
M. DEMURU, F. FARA & M. FRASCHINI
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
this brain rhythm has been previously linked to both imagery and actual movements (McFarland et al., 2000). In conclusion, this paper shows that brain network analysis of EEG functional connectivity could represent an e±cient alternative to more classical local activation description methods, suggesting a shift toward methods based on the characterization of a limited set of functional connections that disclose salient network topological features. In the future, it may be particularly relevant to investigate methods in order to add more time-resolution (particularly relevant in BCI applications) for instance studying time-varying graphs and to train classi¯ers to test the accuracy of a system based on the reported network characteristics. Acknowledgments M. Fraschini was in part supported by the Fondazione Banco di Sardegna (Prot. U7989.2013/AI.713.MGB Prat.2013.1237). M. Demuru was supported by Master and Back Program 2010/2011–Regione Autonoma della Sardegna. REFERENCES Boersma, M., Smit, D.J.A., Boomsma, D.I., De Geus, E.J.C., Delemarre-van de Waal, H.A. & Stam, C.J. (2013) Growing trees in child brains: Graph theoretical analysis of electroencephalography-derived minimum spanning tree in 5- and 7-year-old children re°ects brain maturation. Brain Connect., 3, 50–60. Brunner, C., Scherer, R., Graimann, B., Supp, G. & Pfurtscheller, G. (2006) Online control of a brain-computer interface using phase synchronization. IEEE Trans. Biomed. Eng., 53, 2501–2506. Bullmore, E. & Sporns, O. (2009) Complex brain networks: Graph theoretical analysis of structural and functional systems. Nat. Rev. Neurosci., 10, 186–198. Daly, I., Nasuto, S.J. & Warwick, K. (2012) Brain computer interface control via functional connectivity dynamics. Pattern Recogn., 45, 2123–2136. Goldberger, A.L., Amaral, L.A.N., Glass, L., Hausdor®, J.M., Ivanov, P.C., Mark, R.G., Mietus, J.E., Moody, G.B., Peng, C.-K. & Stanley, H.E. (2000) PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation, 101, e215–e220. Holper, L., Muehlemann, T., Scholkmann, F., Eng, K., Kiper, D. & Wolf, M. (2010) Testing the potential of a virtual reality neurorehabilitation system during performance of observation, imagery and imitation of motor actions recorded by wireless functional nearinfrared spectroscopy (fNIRS). J. Neuroeng. Rehabil., 7, 57. Janssen, L., Meulenbroek, R.G.J. & Steenbergen, B. (2011) Behavioral evidence for lefthemisphere specialization of motor planning. Experimental brain research. Experimentelle Hirnforschung. Experimentation Cerebrale, 209, 65–72. Makeig, S., Kothe, C., Mullen, T., Bigdely-Shamlo, N. & Kreutz-Delgado, K. (2012) Evolving signal processing for brain–computer interfaces. Proc. IEEE, 100, 1567–1584. McFarland, D.J., Miner, L.A., Vaughan, T.M. & Wolpaw, J.R. (2000) Mu and beta rhythm topographies during motor imagery and actual movements. Brain Topogr., 12, 177–186.
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
BRAIN NETWORK ANALYSIS OF EEG FUNCTIONAL CONNECTIVITY
447
Mulder, T. (2007) Motor imagery and action observation: Cognitive tools for rehabilitation. J. Neural Transm., 114, 1265–1278. Nicolas-Alonso, L.F. & Gomez-Gil, J. (2012) Brain computer interfaces, a review. Sensors, 12, 1211–1279. Poznanski, R.R. (2009) Model-based neuroimaging for cognitive computing. J. Integr. Neurosci., 8, 345–369. Roy, S. (2012) Systems biology beyond degree, hubs and scale-free networks: The case for multiple metrics in complex networks. Syst. Synthetic Biol., 6, 31–34. Schalk, G., McFarland, D.J., Hinterberger, T., Birbaumer, N. & Wolpaw, J.R. (2004) BCI2000: A general-purpose brain-computer interface (BCI) system. IEEE Trans. Biomed. Eng., 51, 1034–1043. Stam, C.J., Nolte, G. & Da®ertshofer, A. (2007) Phase lag index: Assessment of functional connectivity from multi channel EEG and MEG with diminished bias from common sources. Hum. Brain Map., 28, 1178–1193. Stam, C.J. & van Straaten, E.C.W. (2012) The organization of physiological brain networks. Clin. Neurophysiol, 123, 1067–1087. Tangermann, M., Müller, K.-R., Aertsen, A., Birbaumer, N., Braun, C., Brunner, C., Leeb, R., Mehring, C., Miller, K.J., Müller-Putz, G.R., Nolte, G., Pfurtscheller, G., Preissl, H., Schalk, G., Schl€ ogl, A., Vidaurre, C., Waldert, S. & Blankertz, B. (2012) Review of the BCI Competition IV. Front. Neurosci., 6, 55. Van Wijk, B.C.M., Stam, C.J. & Da®ertshofer, A. (2010) Comparing brain networks of di®erent size and connectivity density using graph theory. PloS one, 5, e13701.
Brain network analysis of EEG functional connectivity during imagery hand movements
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
Matteo Demuru, Francesca Fara and Matteo Fraschini* Dipartimento di Ingegneria Elettrica ed Elettronica Universit a di Cagliari, Italia *[email protected] [Received 18 July 2013; Accepted 9 September 2013; Published 25 October 2013] The characterization of human neural activity during imaginary movement tasks represent an important challenge in order to develop e®ective applications that allow the control of a machine. Yet methods based on brain network analysis of functional connectivity have been scarcely investigated. As a result we use graph theoretic methods to investigate the functional connectivity and brain network measures in order to characterize imagery hand movements in a set of healthy subjects. The results of the present study show that functional connectivity analysis and minimum spanning tree (MST) parameters allow to successfully discriminate between imagery hand movements (both right and left) and resting state conditions. In conclusion, this paper shows that brain network analysis of EEG functional connectivity could represent an e±cient alternative to more classical local activation based approaches. Furthermore, it also suggests the shift toward methods based on the characterization of a limited set of fundamental functional connections that disclose salient network topological features. Keywords: Imagery movements; functional connectivity; brain networks; minimum spanning tree; EEG; beta band.
1. Introduction The characterization and description of human neural activity during imaginary movement tasks still represent an important challenge. Currently, great interest about both the neuro-rehabilitation techniques based on motor imagery (Mulder, 2007; Holper et al., 2010) and brain computer interface (BCI) systems based on the interpretation of brain activity, attempt to provide (partially or completely) paralyzed patients with a potentially new communication tool. Some very recent reviews (Makeig et al., 2012; Nicolas-Alonso & Gomez-Gil, 2012; Tangermann et al., 2012) outlined the state-of-the-art of BCI systems reporting and discussing the commonly used methods for the characterization of the recorded neurological activity. Methods based on brain network analysis of functional connectivity have been scarcely investigated (Brunner et al., 2006; Daly et al., 2012). These graph theoretical methods have been widely adopted to successfully describe human brain dynamics (Bullmore & Sporns, 2009; *Corresponding
author.
441
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
442
M. DEMURU, F. FARA & M. FRASCHINI
Stam & van Straaten, 2012). Although brain network approach represents an important tool that could be used to develop more e±cient BCI systems, it may not be su±cient to thoroughly describe human brain dynamics as just information °ow in a complex anatomical and functional network (Poznanski, 2009). The aim of the present paper is to investigate the potentiality of functional connectivity analysis and brain network measures in order to characterize imagery (right and left) hand movements in a set of healthy subjects. In particular, we aim to understand if this approach could successfully allow to discriminate between imagery right and left hand movements, and between imagery movements (both right and left) and resting state condition. Functional connectivity was estimated by using a well-known measure of phase synchronization, named phase lag index (PLI) (Stam et al., 2007), which also allowed controlling the relevant problem of volume conduction. Typical normalized clustering coe±cient and shortest path length were evaluated in order to describe the topology of the brain networks, and the use of the minimum spanning tree (MST) (Boersma et al., 2013) has been proposed as a characteristic functional backbone of the network and as a possible solution to unbiased network comparison. 2. Materials and Methods The dataset used in this paper, public and freely downloadable from the web (http:// physionet.org/pn4/eegmmidb/), was created and contributed by the developers of the BCI2000 instrumentation system (Goldberger et al., 2000; Schalk et al., 2004). The dataset consists of 64 channels EEG recordings obtained from 109 volunteers and is provided in EDFþ format. Subjects performed di®erent motor/imagery tasks while 64 channels EEG signals were recorded at 160 Hz sampling rate. Each subject performed 14 experimental runs: two one-minute baseline runs (one with eyes open, one with eyes closed), and three two-minute runs of each of the four following tasks: (i) the subject opens and closes ¯st; (ii) the subject imagines opening and closing the ¯st; (iii) the subject opens and closes either both ¯sts or both feet; (iv) the subject imagines opening and closing either both ¯sts or both feet. Each of the four described task was organized in multiple trials of 4.1 s alternating same length resting state trials. In our analysis we studied the task (ii), so taking into account only imagined right and left hand movement trials and resting state trials between the tasks. Eighteen trials (4,1 s long) were analyzed for each subject and condition (right hand, left hand, resting condition before right hand and resting condition before left hand). The signal analysis was based on estimating the phase synchronization of all pairs of sensors during each trial for each of the 105 subjects (four subjects were successively excluded from analysis because recording parameters were di®erent from the other subjects). The data was averaged over all trials of the same task and averaged over each repetition.
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
BRAIN NETWORK ANALYSIS OF EEG FUNCTIONAL CONNECTIVITY
443
The described analysis was performed for each of the following frequency bands: delta (0.5–3.5 Hz), theta (4–8 Hz), alpha (8–13 Hz) and beta (13–30 Hz). To evaluate the pair-wise phase coupling between EEG signals we estimated the corresponding PLI (Stam et al., 2007). PLI re°ects the consistency with which one signal is phase leading or lagging with respect to another signal, allowing evaluating the phase di®erences between two signals with a measure not a®ected by the in°uence of volume conduction. Besides a global approach (mean PLI calculation) a more regional approach was also successively used. EEG channels were grouped into each hemisphere, and average PLI for all channels within a hemisphere (right and left) were computed. From synchronization matrices, we analyzed fully weighted and undirected graphs by means of the clustering coe±cient (which denotes the likelihood that neighbors of a vertex will also be connected to each other) and the characteristic path length (i.e., the average number of edges of the shortest path between pairs of vertices). All metrics were successively normalized by using randomly rewired versions of the original networks (averaged over 50 repetitions) matched for degree distribution. Furthermore, from every connectivity matrix, the MST was computed (by Kruskal's algorithm) and from these MST graphs several measures were successively estimated: diameter (largest distance between any two nodes), eccentricity (longest distance between a node and any other node), leaf number (number of nodes with degree ¼ 1) and hierarchy (refers to balance in hub nodes). As already reported (Boersma et al., 2013), these measures allow describing and characterizing the topology of the related MST graphs. All the analysis was performed using the Brainwave software (version 0.9.70, http://home.kpn.nl/stam7883/brainwave.html). Statistical analysis was performed with PRISM 5 for Mac OS X version 5.0f (GraphPad Software, Inc.). Two-way ANOVA for repeated measures, using condition (imagery movement/resting state) and side (left/right hand) as within-subject factors, was used to test di®erences for each of the computed measures. Two-way ANOVA for repeated measures, using region (right/left hemisphere) and side (left/ right hand) as within-subject factors, was used to test di®erences for regional (hemispheric) synchronization. A signi¯cance level of 0.05 was used for the test. Successively, Sidak's multiple comparisons test was used to correct for multiple comparisons. 3. Results Global synchronization analysis (mean PLI over all channels) showed a signi¯cant e®ect of condition (imagery movement/resting state) (F½1; 104 ¼ 23:10, p < 0:0001) but no e®ect of side (left/right hand) in the beta band. Sidak's multiple comparisons test showed a signi¯cant global synchronization increase both for right (tð104Þ ¼ 5:344, p < 0:0001) and left (tð104Þ ¼ 3:755, p < 0:0001) imagery hand movement when compared with respective resting state conditions in the beta band (Table 1).
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
444
M. DEMURU, F. FARA & M. FRASCHINI
Regional synchronization analysis (mean PLI over all channels within hemisphere) showed a signi¯cant side per region interaction e®ect (F½1; 104 ¼ 20:47, p < 0:0001) and a signi¯cant regional e®ect (F½1; 104 ¼ 144:0, p < 0:0001) in the beta band. Sidak's multiple comparisons test showed a signi¯cant regional synchronization di®erence in the left hemisphere between right and left hand imagery movement (tð104Þ ¼ 4:398, p < 0:0001) in the beta band (Table 2). No variations for both global and regional synchronization analysis were found in the other frequency bands. No signi¯cant e®ects of condition (imagery movement/resting state) and side (left/right hand) were observed for both normalized clustering coe±cient and characteristic path length in any frequency band. Signi¯cant e®ect of condition (imagery movement/resting state) and side (left/ right hand) were observed for MST eccentricity (F½1; 104 ¼ 41:58, p < 0:0001 and F½1; 104 ¼ 7:345, p ¼ 0:0079, respectively) and MST diameter (F½1; 104 ¼ 40:92, p < 0:0001 and F½1; 104 ¼ 8:148, p ¼ 0:0052, respectively) in the beta band (see Fig. 1). Signi¯cant e®ect of condition (imagery movement/resting state) was observed for MST hierarchy (F½1; 104 ¼ 4:793, p ¼ 0:31) in the beta band. No significant e®ects of condition (imagery movement/resting state) and side (left/right hand) were observed for MST leaf in any frequency band.
Table 1. Results of global PLI in beta band. Statistics refers to post test. IM refers to imagery hand movement condition, RS refers to resting state condition. Left hd refers to left hand and Right hd refers to right hand. Condition
Side
Left hd Right hd
IM
RS
Statisitics
0.185 0.028 0.186 0.028
0.182 0.027 0.182 0.027
*** ****
***p > 0:0001. ****p < 0:00001.
Table 2. Results of regional PLI in beta band. Statistics refers to post test (NS for not signi¯cant). Left hm refers to left hemisphere region, Right hm refers to right hemisphere region. Left hd refers to left hand and Right hd refers to right hand. Side
Region
Left hm Right hm
***p < 0:0001.
Left hd
Right hd
Statistics
0.218 0.038 0.194 0.030
0.222 0.036 0.192 0.029
*** NS
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
BRAIN NETWORK ANALYSIS OF EEG FUNCTIONAL CONNECTIVITY
445
Fig. 1. Signi¯cant repeated measures ANOVA e®ects for MST parameters in beta band. Left side shows MST diameter values for condition and imagery movement hand side. Right side shows MST eccentricity values for condition and imagery movement hand side. Values are expressed as mean and standard deviation.
No di®erences between condition and side for any other network measures were observed. 4. Discussion The results of the present study show that functional connectivity analysis should be considered as an e±cient tool in order to characterize brain dynamics involved during imagery hand movement. Indeed, it has been successfully reported that phase synchronization analysis optimally allowed to discriminate between imagery hand movements (both right and left) and resting state conditions. Moreover, it seems that the left hemisphere regional synchronization mechanisms could play a more relevant role to discriminate between right and left hand movements. This ¯nding, which is not unexpected (Janssen et al., 2011), represents, in our opinion, probably the main result of this study and potentially also suggests a preferential area to be investigated during BCI and neuro-rehabilitation applications. An important issue that should be always addressed when dealing with EEG systems is related to volume conduction problem. Even if our approach (PLI) is particularly conservative, since it discards also possible real zero-lag synchronization, our results still show that synchronization mechanism can be used to discriminate EEG brain dynamics. Furthermore, brain network analysis, in particular the MST parameters characterization, showed that it is possible to observe network topology modi¯cations induced by imagery hand movements compared to resting state conditions. For the ¯rst time, to the best of our knowledge, in the present paper the problem to characterize brain dynamics related to imagery hand movement using a large set of brain network measures (Roy, 2012) and also (using the MST graphs) the overlooked relevant problem of network comparison has been addressed (Van Wijk et al., 2010). Our results strongly suggest that these parameters could be used to discriminate imagery movements from resting state conditions and that the MST could identify a functional backbone to be preferentially investigated in any application. It also should not surprise that our results were observed in the beta (13–30 Hz) band, since
446
M. DEMURU, F. FARA & M. FRASCHINI
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
this brain rhythm has been previously linked to both imagery and actual movements (McFarland et al., 2000). In conclusion, this paper shows that brain network analysis of EEG functional connectivity could represent an e±cient alternative to more classical local activation description methods, suggesting a shift toward methods based on the characterization of a limited set of functional connections that disclose salient network topological features. In the future, it may be particularly relevant to investigate methods in order to add more time-resolution (particularly relevant in BCI applications) for instance studying time-varying graphs and to train classi¯ers to test the accuracy of a system based on the reported network characteristics. Acknowledgments M. Fraschini was in part supported by the Fondazione Banco di Sardegna (Prot. U7989.2013/AI.713.MGB Prat.2013.1237). M. Demuru was supported by Master and Back Program 2010/2011–Regione Autonoma della Sardegna. REFERENCES Boersma, M., Smit, D.J.A., Boomsma, D.I., De Geus, E.J.C., Delemarre-van de Waal, H.A. & Stam, C.J. (2013) Growing trees in child brains: Graph theoretical analysis of electroencephalography-derived minimum spanning tree in 5- and 7-year-old children re°ects brain maturation. Brain Connect., 3, 50–60. Brunner, C., Scherer, R., Graimann, B., Supp, G. & Pfurtscheller, G. (2006) Online control of a brain-computer interface using phase synchronization. IEEE Trans. Biomed. Eng., 53, 2501–2506. Bullmore, E. & Sporns, O. (2009) Complex brain networks: Graph theoretical analysis of structural and functional systems. Nat. Rev. Neurosci., 10, 186–198. Daly, I., Nasuto, S.J. & Warwick, K. (2012) Brain computer interface control via functional connectivity dynamics. Pattern Recogn., 45, 2123–2136. Goldberger, A.L., Amaral, L.A.N., Glass, L., Hausdor®, J.M., Ivanov, P.C., Mark, R.G., Mietus, J.E., Moody, G.B., Peng, C.-K. & Stanley, H.E. (2000) PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation, 101, e215–e220. Holper, L., Muehlemann, T., Scholkmann, F., Eng, K., Kiper, D. & Wolf, M. (2010) Testing the potential of a virtual reality neurorehabilitation system during performance of observation, imagery and imitation of motor actions recorded by wireless functional nearinfrared spectroscopy (fNIRS). J. Neuroeng. Rehabil., 7, 57. Janssen, L., Meulenbroek, R.G.J. & Steenbergen, B. (2011) Behavioral evidence for lefthemisphere specialization of motor planning. Experimental brain research. Experimentelle Hirnforschung. Experimentation Cerebrale, 209, 65–72. Makeig, S., Kothe, C., Mullen, T., Bigdely-Shamlo, N. & Kreutz-Delgado, K. (2012) Evolving signal processing for brain–computer interfaces. Proc. IEEE, 100, 1567–1584. McFarland, D.J., Miner, L.A., Vaughan, T.M. & Wolpaw, J.R. (2000) Mu and beta rhythm topographies during motor imagery and actual movements. Brain Topogr., 12, 177–186.
J. Integr. Neurosci. 2013.12:441-447. Downloaded from www.worldscientific.com by UNIVERSIT OF SOUTHERN CALIFORNIA on 11/13/14. For personal use only.
BRAIN NETWORK ANALYSIS OF EEG FUNCTIONAL CONNECTIVITY
447
Mulder, T. (2007) Motor imagery and action observation: Cognitive tools for rehabilitation. J. Neural Transm., 114, 1265–1278. Nicolas-Alonso, L.F. & Gomez-Gil, J. (2012) Brain computer interfaces, a review. Sensors, 12, 1211–1279. Poznanski, R.R. (2009) Model-based neuroimaging for cognitive computing. J. Integr. Neurosci., 8, 345–369. Roy, S. (2012) Systems biology beyond degree, hubs and scale-free networks: The case for multiple metrics in complex networks. Syst. Synthetic Biol., 6, 31–34. Schalk, G., McFarland, D.J., Hinterberger, T., Birbaumer, N. & Wolpaw, J.R. (2004) BCI2000: A general-purpose brain-computer interface (BCI) system. IEEE Trans. Biomed. Eng., 51, 1034–1043. Stam, C.J., Nolte, G. & Da®ertshofer, A. (2007) Phase lag index: Assessment of functional connectivity from multi channel EEG and MEG with diminished bias from common sources. Hum. Brain Map., 28, 1178–1193. Stam, C.J. & van Straaten, E.C.W. (2012) The organization of physiological brain networks. Clin. Neurophysiol, 123, 1067–1087. Tangermann, M., Müller, K.-R., Aertsen, A., Birbaumer, N., Braun, C., Brunner, C., Leeb, R., Mehring, C., Miller, K.J., Müller-Putz, G.R., Nolte, G., Pfurtscheller, G., Preissl, H., Schalk, G., Schl€ ogl, A., Vidaurre, C., Waldert, S. & Blankertz, B. (2012) Review of the BCI Competition IV. Front. Neurosci., 6, 55. Van Wijk, B.C.M., Stam, C.J. & Da®ertshofer, A. (2010) Comparing brain networks of di®erent size and connectivity density using graph theory. PloS one, 5, e13701.
