Predicting neurosurgical outcomes in focal epilepsy patients using computational modelling

Abstract

SEE EISSA AND SCHEVON DOI101093/AWW332 FOR A SCIENTIFIC COMMENTARY ON THIS ARTICLE: Surgery can be a last resort for patients with intractable, medically refractory epilepsy. For many of these patients, however, there is substantial risk that the surgery will be ineffective. The prediction of who is likely to benefit from a surgical approach is crucial for being able to inform patients better, conduct principled prospective clinical trials, and ultimately tailor therapeutic approaches to these patients more effectively. Dynamical computational models, informed with patient data, can be used to make predictions and give mechanistic insight. In this study, we develop patient-specific dynamical network models of epileptogenic cortex. We infer the network connectivity matrix from non-seizure electrographic recordings of patients and use these connectivity matrices as the network structure in our model. The model simulates the dynamics of a bi-stable switch at every node in this network, meaning that every node starts in a background state, but has the ability to transit to a co-existing seizure state. Whether a transition happens in a node is partly determined by the stochastic nature of the input to the node, but also by the input the node receives from other connected nodes in the network. By conducting simulations with such a model, we can detect the average transition time for nodes in a given network, and therefore define nodes with a short transition time as highly epileptogenic. In a retrospective study, we found that in some patients the regions with high epileptogenicity in the model overlap with those identified clinically as the seizure onset zone. Moreover, it was found that the resection of these regions in the model reduces the overall likelihood of a seizure. Following removal of these regions in the model, we predicted surgical outcomes and compared these to actual patient outcomes. Our predictions were found to be 81.3% accurate on a dataset of 16 patients with intractable epilepsy. Intriguingly, in patients with unsuccessful outcomes, the proposed computational approach is able to suggest alternative resection sites. The model presented here gives mechanistic insight as to why surgery may be unsuccessful in some patients. This may aid clinicians in presurgical evaluation by providing a tool to explore various surgical options, offering complementary information to existing clinical techniques.

Keywords: computational models; epilepsy; focal seizures; intracranial EEG; surgical outcome prediction.

Figure 1

Illustration of model dynamics. (A) Deterministic dynamics of a single node representing the bi-stability of the model. (B) Stochastic dynamics in a two node network. The two nodes are initially disconnected having independent dynamics. Depending on the strength and direction of connections, the dynamics of each node is influenced by the other. (C and D) Patient-specific connectivity matrix is obtained from intracranial, interictal ECoG recording, which is incorporated as a model parameter to simulate the model dynamics.

Figure 2

Illustration of seizure likelihood computation. (A) Electrodes in seizure onset zone (4, 5, 6, 10, 11, 12, 16, 17, 18) are shown in red on the brain schematic. The connectivity matrix inferred from the ECoG recordings is coupled with the model and the model dynamics is simulated with 1000 different noise realizations (B) The bar graph represents the seizure likelihood for each node and the error bars represent the standard error. Note that the nodes with significantly higher seizure likelihood (indicated by an asterisk) are correlated with the seizure onset zone shown in red in A and B.

Figure 3

Overall procedure. (A–C) The computation of functional connectivity by averaging the windowed correlation matrices estimated from the segmented interictal ECoG signals. We coupled the model with the modified connectivity matrix from step D to compute the seizure likelihood upon actual resection (as shown in step H). Similarly, we computed seizure likelihood upon random resection (illustrated in step I) by coupling the modified connectivity matrix from step E with the model. From the steps H and I, we made predictions about surgical outcome by comparing their efficacy on seizure reduction in the model.

Figure 4

Correlation between clinical resection, post-surgical outcome and seizure likelihood. Cortical areas under electrode channels which were surgically resected have been shaded in black. Post-surgery, Patient P1 shown in A had a good surgical outcome (ILAE class II); while Patient P2 in B had a poor surgical outcome (ILAE class IV). The colour plot on which the electrodes are overlaid shows the distribution of simulated seizure likelihood values of different brain regions.

Figure 5

Node removal to predict surgical outcome. Resected cortical tissues are coloured in red. Nodes within the resected tissue are removed from the model. The resulting increase in escape time is shown in the box plot (in red), which is compared against the increase in escape time due to removal of the same number of randomly selected nodes, averaged over 100 instances (in blue). *P = 0.005–0.05; **P = 0.0005–0.005; ***P < 0.0005 computed using the non-parametric Wilcoxon rank sum test.

Figure 6

Illustrating in silico approach for exploring surgical options. The seizure likelihood for each ECoG channel is shown in the bar plot. Higher seizure likelihood indicates more propensity to seize. Nodes with significantly higher seizure likelihood after FDR correction are indicated in red in the bar plot and their spatial locations are mapped on the electrode grids in black. Nodes are removed in the model brain to simulate surgical resection. The box plots show escape time for (i) original network (in green); (ii) resection of nodes with the highest seizure likelihood (in red); and (iii) resection of same number of random nodes, averaged over 100 instances (in blue). ***P < 0.0005.