Individual brain structure and modelling predict seizure propagation

Abstract

See Lytton (doi:10.1093/awx018) for a scientific commentary on this article.Neural network oscillations are a fundamental mechanism for cognition, perception and consciousness. Consequently, perturbations of network activity play an important role in the pathophysiology of brain disorders. When structural information from non-invasive brain imaging is merged with mathematical modelling, then generative brain network models constitute personalized in silico platforms for the exploration of causal mechanisms of brain function and clinical hypothesis testing. We here demonstrate with the example of drug-resistant epilepsy that patient-specific virtual brain models derived from diffusion magnetic resonance imaging have sufficient predictive power to improve diagnosis and surgery outcome. In partial epilepsy, seizures originate in a local network, the so-called epileptogenic zone, before recruiting other close or distant brain regions. We create personalized large-scale brain networks for 15 patients and simulate the individual seizure propagation patterns. Model validation is performed against the presurgical stereotactic electroencephalography data and the standard-of-care clinical evaluation. We demonstrate that the individual brain models account for the patient seizure propagation patterns, explain the variability in postsurgical success, but do not reliably augment with the use of patient-specific connectivity. Our results show that connectome-based brain network models have the capacity to explain changes in the organization of brain activity as observed in some brain disorders, thus opening up avenues towards discovery of novel clinical interventions.

Keywords: brain network models; connectomes; epilepsy; individualized medicine; seizure propagation.

Figure 1

Stereotactic EEG data and reconstruction of virtual Patient CJ. (A) Two examples of partial seizures recorded with stereotactic EEG in this patient. Left, the seizure propagates from the epileptogenic zone (yellow) to the propagation zone (red). Right, the seizure is limited to the epileptogenic zone. The normalized energy of each channel is shown in colour on the side, with below the corresponding colour bar. Spatiotemporal activation patterns are shown at different time points of the seizures. lSPC = left superior parietal cortex; lIPC = left inferior parietal cortex; lLgG = left lingual gyrus; lLOC1 = part 1 of the left lateral occipital cortex; lFuG = left fusiform gyrus; lLOC2 = part 2 of the left lateral occipital cortex; lITG = left inferior temporal gyrus. (B) Coregistration of the T₁ MRI (levels of grey), the parcellation with 157 regions (colours) and the intracranial electrodes (red strips). (C) Connectivity matrix obtained from diffusion MRI for this parcellation. Ls = left subcortical; Rs = right subcortical.

Figure 2

Simulations of the BNM for the Patient CJ. (A) A network of Epileptor models is build using the connectivity matrix. The nodes in the epileptogenic zone are epileptogenic (Δx_0,i > 0, in yellow), the nodes in the propagation zone have different excitability values (0.5 > Δx_0,i> −0.5, shades of red), while all the other nodes are not epileptogenic (stable state, Δx_0,i< 0, in shades of blue). The blue links represent the anatomical links of the connectivity matrix. (B) Example of time series generated by the simulated BNM with the connectome of Patient CJ. Without changing any parameters, the propagation zone is not always recruited, reproducing the two seizures types of this patient as shown in Fig. 1A.

Figure 3

Prediction of the propagation zone by linear stability analysis for Patient CJ. (A) A network of Epileptor models is built using the connectivity matrix. The nodes in the epileptogenic zone (EZ) are epileptogenic (Δx_0,i< 0, in yellow), while all the other nodes are equally far from the epileptogenicity threshold (Δx_0,i< −0.5, in blue). Examples of the localization of the epileptogenic zone (yellow) and the propagation zone (PZ) (shade of red, as indicated by the colour bar) in Patient CJ such as found by: (B) linear stability analysis using the patient connectome; (C) stereotactic EEG clinician estimations; (D) stereotactic EEG signal quantifications. The propagation zone values are all the same for the stereotactic EEG clinician estimations. SEEG = stereotactic EEG.

Figure 4

Prediction of the propagation zone by linear stability analysis for Patient GC. A network of Epileptor models is built using the connectivity matrix. The nodes in the epileptogenic zone (EZ) are epileptogenic (Δx_0,i< 0, in yellow), while all the other nodes are equally far from the epileptogenicity threshold (Δx_0,i< −0.5, in blue). Examples of the localization of the epileptogenic zone (yellow) and the propagation zone (PZ) (shade of red, as indicated by the colour bar) in Patient CJ such as found by: (B) linear stability analysis using the patient connectome; (C) stereotactic EEG clinician estimations; (D) stereotactic EEG signal quantifications. The propagation zone values are all the same for the stereotactic EEG clinician estimations. SEEG = stereotactic EEG.

Figure 5

Prediction of the propagation zone by linear stability analysis for Patient ML. A network of Epileptor models is build using the connectivity matrix. The nodes in the epileptogenic zone (EZ) are epileptogenic (Δx_0,i< 0, in yellow), while all the other nodes are equally far from the epileptogenicity threshold (Δx_0,i< −0.5, in blue). Examples of the localization of the epileptogenic zone (yellow) and the propagation zone (PZ) (shade of red, as indicated by the colour bar) in Patient CJ such as found by: (B) linear stability analysis using the patient connectome; (C) stereotactic EEG clinician estimations; (D) stereotactic EEG signal quantifications. The propagation zone values are all the same for the stereotactic EEG clinician estimations. SEEG = stereotactic EEG.

Figure 6

Prediction scores for patient, control subject and shuffled connectivity matrices. Results for Patients CJ, GC and ML, compared to five control subjects and to shuffled connectivity matrix, for propagation zone (PZ) location according to (A) stereotactic EEG clinician estimation, and (B) stereotactic EEG signal quantification. The dashed line indicates the level of chance. In box plots, boxes represent the 25th and 75th percentiles, centre line indicates the median and the whiskers extend to the most extreme data points not considered outliers. SEEG = stereotactic EEG.

Figure 7

Prediction of the surgical outcome. (A) Example for Patient CJ showing stereotactic EEG signal quantification of the propagation zone (PZ) overlaid in green by the regions found in the propagation zone by the linear stability analysis but not explored by stereotactic EEG, therefore not considered in the propagation zone by the clinical expertise. (B) Comparison for all patients of the size of the unexplored regions predicted as in the propagation zone by the analytical model and the Engel classification (stereotactic EEG clinician estimation, 158 regions, distance score). Significant P-values are indicated. ^**P < 0.05, ^*P < 0.1; Mann-Whitney U-test.