Tackling the challenges of group network inference from intracranial EEG data
Poster No:
1546
Submission Type:
Abstract Submission
Authors:
Anna Pidnebesna1, Pavel Sanda1, Adam Kalina2, Jiri Hammer2, Petr Marusic2, Kamil Vlcek3, Jaroslav Hlinka1
Institutions:
1Institute of Computer Science of the Czech Academy of Sciences, Prague, Czech Republic, 2Charles University, 2nd School of Medicine, University Hospital Motol, Prague, Czech Republic, 3Institute of Physiology of the Czech Academy of Sciences, Prague, Czech Republic
First Author:
Anna Pidnebesna
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic
Co-Author(s):
Adam Kalina
Charles University, 2nd School of Medicine, University Hospital Motol
Prague, Czech Republic
Charles University, 2nd School of Medicine, University Hospital Motol
Prague, Czech Republic
Jiri Hammer
Charles University, 2nd School of Medicine, University Hospital Motol
Prague, Czech Republic
Charles University, 2nd School of Medicine, University Hospital Motol
Prague, Czech Republic
Petr Marusic
Charles University, 2nd School of Medicine, University Hospital Motol
Prague, Czech Republic
Charles University, 2nd School of Medicine, University Hospital Motol
Prague, Czech Republic
Jaroslav Hlinka
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic
Institute of Computer Science of the Czech Academy of Sciences
Prague, Czech Republic
Introduction:
Effective connectivity, or the causal relationships between brain areas, is an important tool for understanding information flow in the brain. However, it can be difficult to estimate accurately due to various challenges. In this study, we address these challenges specifically in inferring group networks from intracranial EEG recordings of epileptic patients during a visual task. These challenges include low and heterogeneous brain coverage, nonlinear signals, the influence of unobserved variables, a limited number of patients, a large amount of time-series data, and difficulties in statistical inference. We present a group network estimation pipeline that deals with these difficulties [1].
Methods:
Data. iEEG was recorded in 15 patients with a sampling frequency of 512 Hz. Recorded time series were normalised to their pre-stimulus standard deviation. Bipolar montage was used to reduce the influence of distant sources. For the connectivity network exploration, 7 ROI were identified as parts of ventral/dorsal pathways [2].
Task. A set of pictures was presented to the subjects on a computer monitor. Brain responses to three types of images were studied (scenes, faces, objects) with 200 trials of each type. Every trial included 200 ms of baseline, 300 ms of stimulus, and 600 ms of reaction time. The trials were aligned on stimulus presentation yielding the event-related potentials (ERPs). Only those contacts with significantly more pronounced responses to scenes than objects were analysed.
Methods. We present a pipeline for group network estimation using an example of a Directed Transfer Function (DTF). We propose to compute the dynamical connectivity for a filtered signal for a set of frequency bands, evaluating the statistical significance via the multivariate Fourier surrogates. Further, binary maps of significant connectivity values are collected among all patients to a heatmap per pair of ROIs. To evaluate the connectivity presence, we compute an average entry of a heatmap for a reaction time, corrected to the average heatmap entry during the baseline. This value is compared to the distribution obtained from the surrogates; FWE correction (p<0.05) is applied in the last step of the analysis (Fig.1).
Task. A set of pictures was presented to the subjects on a computer monitor. Brain responses to three types of images were studied (scenes, faces, objects) with 200 trials of each type. Every trial included 200 ms of baseline, 300 ms of stimulus, and 600 ms of reaction time. The trials were aligned on stimulus presentation yielding the event-related potentials (ERPs). Only those contacts with significantly more pronounced responses to scenes than objects were analysed.
Methods. We present a pipeline for group network estimation using an example of a Directed Transfer Function (DTF). We propose to compute the dynamical connectivity for a filtered signal for a set of frequency bands, evaluating the statistical significance via the multivariate Fourier surrogates. Further, binary maps of significant connectivity values are collected among all patients to a heatmap per pair of ROIs. To evaluate the connectivity presence, we compute an average entry of a heatmap for a reaction time, corrected to the average heatmap entry during the baseline. This value is compared to the distribution obtained from the surrogates; FWE correction (p<0.05) is applied in the last step of the analysis (Fig.1).
·Diagram of the pipeline showing the subsequent steps starting from the time series measured from a single patient, through computation of dynamic connectivity, binary maps, heatmaps and test statisti
Results:
Although the specific application was not the main rationale of this study, the utilisation of the developed pipeline provided interesting results. Our data show that the indirect pathway connecting the parietal lobe with locations in the medial temporal lobe via the retrosplenial complex is active during static visual scene processing in humans.
Conclusions:
Based on the discussion of the possible ways to tackle challenges named in the introduction, we have made a range of methodological decisions to build up a pipeline allowing robust statistical inference of the alterations of the brain connectivity network during processing visual stimuli. The application of the pipeline was demonstrated on an example dataset, giving rise to a group-level network in terms of functional and effective connectivity.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
Novel Imaging Acquisition Methods:
EEG 2
Perception, Attention and Motor Behavior:
Perception: Visual
Keywords:
Data analysis
Multivariate
Statistical Methods
Vision
1|2Indicates the priority used for review
Provide references using author date format
[2] Vlcek, K. (2020). Mapping the Scene and Object Processing Networks by Intracranial EEG. Frontiers in Human Neuroscience, 14. https://doi.org/10.3389/fnhum.2020.561399