Evaluation of statistical significance of effective connectivity measures in ECoG signals
Abstract number :
1.149
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2017
Submission ID :
346060
Source :
www.aesnet.org
Presentation date :
12/2/2017 5:02:24 PM
Published date :
Nov 20, 2017, 11:02 AM
Authors :
Ali Moharramipour, School of ECE, College of Engineering, University of Tehran, Iran; Parham Mostame, School of ECE, College of Engineering, University of Tehran, Iran; Gholam- Ali Hossein-Zadeh, School of ECE, College of Engineering, University of Tehran
Rationale: Effective connectivity determines causal interactions between brain regions. A popular method for the effective connectivity analysis is the multivariate autoregressive (MVAR) model. One challenge in this analysis is to identify significant connections. In this study, we evaluated performance of three resampling techniques, i.e. leave one out method (LOOM), bootstrap resampling, and permutation test, for identification of significant connections in electrocorticography (ECoG) data. Methods: ECoG data were collected while patient performed an overt auditory verb generation (AVG) task. After preprocessing and trial selection, the best model order for MVAR was selected based on the Akaike criterion [1]. Partial directed coherence (PDC) was applied on MVAR model to determine the direct causal connections between each pairs of subdural electrodes [2], and significantly connections were identified using three resampling techniques. In LOOM, one trial was excluded once, the PDC analysis was applied to the rest of the trials, and alteration of the connectivity from baseline to after stimulus onset was calculated. After repeating this procedure over all trials, two-sample t-test was applied to find significant connections [3]. We utilized bootstrap with replacement (n = 200) across trials to calculate a histogram for the baseline PDC and then find a threshold for the significance level [4]. In permutation test, surrogate data was generated by randomly permuting phase of ECoG data. We subtracted mean value of PDC of the surrogate data in baseline from its value after stimulus onset. Then we calculated a histogram at each time-frequency bins after stimulus onset, and identified threshold of the significance level. Results: We investigated the brain connectivity in high gamma frequency band in this study. We used p-value < 0.01 and evaluated the number of significant connections over time for three resampling techniques (Table 1). The bootstrap seems to be too conservative as it identified minimum number of significant connections among three methods. Conversely, the LOOM identified maximum number of significant connections, although some of these connections might be false positive. The permutation test provided more reasonable results compared to other two methods. Fig. 1 shows significant connections, identified using the permutation test, during the AVG task. Superior temporal gyrus had significant connections with other areas during perception of nouns (before 1.6 second). The Broca’s area and Motor cortex had significant interactions during articulation of the verbs. Conclusions: Our results revealed that the permutation test outperformed the bootstrap and LOOM, and this test can be used to identify significant causal interactions in ECoG data. Funding: This study was funded by the Children’s Foundation Research Institute, Memphis, TN.
Neurophysiology