computational modelling and electrical stimulation for epilepsy surgery
Gerrit Jan Hebbink is a PhD student in the research group Systems, Analysis and Computational Sciences. His supervisor is prof.dr. S.A. van Gils from the faculty of Electrical Engineering, Mathematics and Computer Science.
Epilepsy is a common neurological disease, affecting around 1% of the world population. Epilepsy is characterized by the occurrence of transient periods of abnormal excessive or synchronous neuronal activity, called epileptic seizures. Epilepsy surgery may provide a cure for patients with focal epilepsy and aims at removing the epileptogenic zone. Accurate delineation of the epileptogenic zone, however, remains one of the challenges in epilepsy surgery.
Traditionally, epilepsy surgery mainly focuses on removing pathological cortex in the anatomic sense, while recent developments suggests that epilepsy is also a brain network disease. Computational models for epilepsy offer a framework to study the joint effect of local, intrinsic epileptogenicity and network interaction and allow to incorporate patient-specific information. In epilepsy patients who undergo long-term invasive EEG recordings prior to surgery, this information might be obtained using single pulse electrical stimulation (SPES). Using brief electrical pulses SPES evokes early responses, representing connectivity, and delayed responses which are a biomarker for epileptogenic cortex. In this thesis the added value of combining computational network models and SPES for epilepsy surgery is investigated.
First, we study the effects of surgery on the seizure rate in a simple network model consisting of four interconnected neuronal populations. One of these populations can be configured to be hyperexcitable, modelling a pathological region of cortex. Model simulations show that removal of normal populations located at a crucial spot in the network, is typically more effective in reducing seizure rate than removing the hyperexcitable node. This result strengthens the idea that besides localizing pathological tissue also network structure must be taken into account for successful epilepsy surgery.
Second, we compare connectivity probed using SPES with two traditional methods, i.e. cross-correlation and Granger causality, that infer connectivity from ongoing brain activity recorded using intracranial EEG. All three methods yield primarily nearest neighbour connections, however SPES networks are usually connected more densely and include more distant connections than cross-correlation and Granger causality networks. We find that strong connections in the cross-correlation network form more or less a subset of the SPES network, while Granger causality and SPES networks are related more weakly. Connectivity known to exist between Broca’s and Wernicke’s area, the two major hubs in the language circuit, is only found in SPES networks. An explanation for the differences between the networks might be that cross-correlation and Granger causality infer connectivity from passive observations, where SPES probes connectivity actively.
Next, we use a data-driven modelling approach to study the mechanism generating delayed responses evoked during SPES. Using data of 11 patients we confirm our hypothesis that delayed responses are indirect responses triggered by early response activity. We show using two feedforward coupled neural mass models that delayed responses can be generated when input to a neural mass falls below a threshold, forcing it into a spiking regime temporarily. The combination of the threshold with noisy background input explains the typical stochastic appearance of delayed responses. The probability for a delayed response to occur, depends on both the intrinsic excitability of a neural mass and the strength of its input. These results gives a theoretical basis for the use of delayed responses as a biomarker for the epileptogenic zone, confirming earlier clinical observations.
The mechanism we propose to model delayed responses is an example of a large non-linear response to a short transient input which appears abruptly while increasing the stimulation strength in an excitable system. Using slow-fast analysis we study the transition from small, more or less linear to large, non-linear responses in more detail in our model for delayed responses. We find that the two response types are separated by a high-dimensional stable manifold of a saddle. Large pathological responses appear if the fast subsystem escapes from this manifold to another attractor. The orbit of the critical response can be formulated as a boundary value problem with one free parameter and can be used to study the dependency of the transition between the two response types upon the system parameters.
Currently, the usage of SPES to assist with localization of the epileptogenic zone is limited to patients undergoing long-term intracranial EEG monitoring. The SPES protocol might be accelerated by exploiting the relation between early and delayed responses. Analysis of delayed responses might be sped-up and improved by using automatic detection based on machine learning. Ultimately, intraoperative use of SPES for precise localization of the epileptogenic zone might be feasible, although still many efforts need to be made.
Computational network models offer an interesting framework to explore effects of epilepsy surgery. Multiple studies have shown that such models in combination with patient-specific networks can predict the outcome of surgery. Networks derived from SPES might be an interesting option to use as these networks incorporate more physiological long-range connections than functional networks derived from ongoing intracranial EEG recordings. An important next step to improve computational network models for epilepsy surgery is to incorporate patient-specific information about local excitability. Our results suggest that delayed responses evoked during SPES might be a good candidate marker for this, with the advantage over other biomarkers that it can be probed actively. In conclusion, SPES and patient-specific computational network models form a promising combination that have great potential to improve epilepsy surgery.
