Introduction

The directed transfer function (DTF) of SEEG functional connectivity in regions throughout the epilepsy network has proved to be capable of detecting the epileptogenic zones and predicting surgical outcomes^1,2. Previous studies have attempted to investigate how structural connection may affect effective functional connectivity^3,4. Further, alterations in metabolic connectivity have been widely reported in epileptic patients⁵. In this study, we investigate whether and how structural connectivity and metabolic connectivity may separately or jointly affect the directional information flow.

Method

Data acquisition: In this IRB-approved study, 15 drug-refractory epilepsy patients with mesial temporal sclerosis were recruited. CT and MRI anatomical images were acquired on routine clinical systems. The SEEG system consisting of 8-16 contacts was used to continuously record patients’ neuronal activities, with a sampling rate of 2kHz. Two to three segments of SEEG recordings of typical ictal events of each patient were extracted for analysis. FDG PET scans were obtained approximately 30-50 minutes post a bolus injection of [18F] FDG, with a mean dose of 6-8mCi. The reconstruction matrix was 1.6×1.6×1.5 mm³, 145 slices, ensuring a comprehensive and detailed representation of brain anatomy and function.
SEEG functional connectivity: Anatomical co-localization of SEEG electrodes was performed by the iEEGview toolbox with preoperative MRI and post-implanted CT images. The brain cortex is parcellated into different regions of interest according to Destrieux atlas^6,7. All SEEG data were band-pass filtered at 1–512 Hz. The event of seizure and the seizure onset zone (SOZ) from SEEG recording were selected by experienced physicians. To study the propagation of the SEEG signal, the pre-ictal stage (1 minute before the seizure onset, and lasts 30 seconds) was extracted for analysis.
The within-frequency directional information flow—directed transfer function (DTF)^8,9 was defined as:
$$DTF(i,j,f)=\frac{|h_{ij}\left(f\right)|}{\sqrt{{\bf h}_j^H\left(f\right){\bf h}_j\left(f\right)}}$$
where $$${\bf h}_j$$$ and $$$h_{ij}$$$ were the Fourier transform of the coefficients of the time series modeled by multivariate autoregressive (MVAR) models. The DTF was then averaged across the frequency bands: delta (1–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), gamma (30–80 Hz), high frequency of ripple (80-250 Hz) and fast ripple (250-500 Hz).
DTI structural connectivity: The HCP-1065 DTI template was applied for white matter connectivity analysis¹⁰. The tractography was performed automatically with DSI-studio (http://dsi-studio.labsolver.org). The anatomical scans were parcellated into 198 ROIs using the Destrieux atlas⁷. The count values of streamlines between two paired ROIs were extracted to generate the structural network for the template brain. For individual analysis, structural nodes in the regions sampled by SEEG electrodes and corresponding connectivity were extracted from the template brain.
PET metabolic network: PET image preprocessing was carried out using the standard statistical parametric mapping (SPM) software (MATLAB 2022b). The SUVRs were obtained via global mean scaling of PET images. The PET images were spatially normalized onto the MNI atlas. Individual t-maps of each patient were extracted by SPM analysis performed in comparison to healthy subjects, which were then inverse transformed back to individual space. The PET metabolic network was constructed by taking difference between brain regions $$$p_{ij}=p_j-p_i$$$.
Statistical tests: A step-wise analysis was performed: 1) explore the dominant direction of information flow between SOZ and non-SOZ; 2) evaluate the impact of structural connectivity on the SEEG information flow; 3) evaluate the associations of inward and outward DTF strength with level of FDG uptake; 4) investigate the associations between DTF strength and joint structural and metabolic connectivity. T-tests were performed for all pair-wise comparisons between SOZ and non-SOZ, and the structural features. Pearson’s correlation was performed for the inward/outward information strength and FDG uptakes, as well as for the DTF information strength and FDG metabolic networks with/without structural connectivity.

Results

SEEG DTF analysis showed dominant information flow from non-SOZ to SOZ in all frequency bands (Fig.1) in pre-ictal state, while the difference is mainly attributed to inward information flow between non-SOZ and SOZ. For the electrodes with/without structural connectivity, the information flows have the same direction from non-SOZ to SOZ (Fig.2). Electrodes with and without structural connectivity had no difference in unidirectional information flow strength. The SEEG inward information flow was negatively correlated with FDG metabolism across patients (P=0.039) in all frequency bands as shown in Fig.3. For the electrodes without structural connectivity, the information flow positively correlated with FDG metabolism networks (P=0.011), which was significantly different with the electrodes with structural connectivity (P=0.019) in all frequency bands (Fig.4).

Conclusion

Our study analyzed the relationship between directed neural information flow and FDG metabolism under different white matter connectivity conditions. These findings highlight the interdependence of functional and structural networks and the metabolism in temporal lobe epilepsy.

Acknowledgements

N/A