Active probing to highlight approaching transitions to ictal states in coupled neural mass models

The extraction of electrophysiological features that reliably forecast the occurrence of seizures is one of the most challenging goals in epilepsy research. Among possible approaches to tackle this problem is the use of active probing paradigms in which responses to stimuli are used to detect underlying system changes leading up to seizures. This work evaluates the theoretical and mechanistic underpinnings of this strategy using two coupled populations of the well-studied Wendling neural mass model. Different model settings are evaluated, shifting parameters (excitability, slow inhibition, or inter-population coupling gains) from normal towards ictal states while probing stimuli are applied every 2 seconds to the input of either one or both populations. The correlation between the extracted features and the ictogenic parameter shifting indicates if the impending transition to the ictal state may be identified in advance. Results show that not only can the response to the probing stimuli forecast seizures but this is true regardless of the altered ictogenic parameter. That is, similar feature changes are highlighted by probing stimuli responses in advance of the seizure including: increased response variance and lag-1 autocorrelation, decreased skewness, and increased mutual information between the outputs of both model subsets. These changes were mostly restricted to the stimulated population, showing a local effect of this perturbational approach. The transition latencies from normal activity to sustained discharges of spikes were not affected, suggesting that stimuli had no pro-ictal effects. However, stimuli were found to elicit interictal-like spikes just before the transition to the ictal state. Furthermore, the observed feature changes highlighted by probing the neuronal populations may reflect the phenomenon of critical slowing down, where increased recovery times from perturbations may signal the loss of a systems’ resilience and are common hallmarks of an impending critical transition. These results provide more evidence that active probing approaches highlight information about underlying system changes involved in ictogenesis and may be able to play a role in assisting seizure forecasting methods which can be incorporated into early-warning systems that ultimately enable closing the loop for targeted seizure-controlling interventions.


In blue: Responses to the issues raised by the editor and reviewers
In red: Changes made to the manuscript (lines refer to the marked-up version).

Reviewer #1:
The article discusses the important issue of predicting epileptic discharges using the existing computational neural mass model. A feature of this work is the analysis of signals received from the model. The task of predicting discharges is solved by modeling external stimuli by changing the parameters of the model.
The disadvantages of the work include: 1.
In the "abstract" section, authors should better identify the methods that were used in this particular work.

Response:
We have specified the model used in the work and the target and rate of the stimulation, in addition to adapting some terms. We tried not to add an overwhelming amount of method details in order to stay within the word limits and keep the abstract clear and concise.
(abstract) …This work evaluates the theoretical and mechanistic underpinnings of this strategy using two coupled populations of the well-studied Wendling neural mass model. Different model settings are evaluated, shifting parameters (excitability, slow inhibition, or inter-population coupling gains) from normal towards ictal states while probing stimuli are applied every 2 seconds to the input of either one or both populations. The correlation between the extracted features and the ictogenic parameter shifting indicates if the impending transition to the ictal state may be identified in advance. …

2.
Using the term lag-1 autocorrelation without clarification at the initial mention.

Response:
We clarified the meaning of the lag-1 autocorrelation measure. Response: It refers to a moving average (moving mean, or rolling mean) filter, commonly used in digital signal processing to smooth noisy time series. The size 20 referred to the order of the filter -that is, each filtered value is taken as the mean of the 20 neighbouring samples. In the text, the phrase was rewritten for clarity.

(Section 2.4, line 223-225)
A moving average filter of order 20 was used for smoothing the resulting feature time series.
Response: It means "arbitrary units", which we initially considered to be the model"s output

(Section 2.4, line 249)
Spikes were detected if a peak was found in the simulated signal with amplitude greater than 5 mV.
( Figure 2 is shown in Comment #6 of Reviewer 2) 5. In my opinion, the "Conclusions" section should be written in a little more detail.

Response:
We have included additional methodological details and emphasized the contributions of our work.

(line 484)
…As parameters of a neuronal population were shifted from normal towards ictal activity, responses showed changes that were in general not visible with the passive observation of the populations" activity, such as increased variance, lag-1 autocorrelation and synchrony between neuronal populations" activities….

(lines 491-493)
… This is often related to the loss of a systems" resilience and are common hallmarks of an impending critical transition…

(Lines 495-496)
Altogether, results indicate that low-frequency probing stimuli can provide predictive value and reveal underlying dynamics involved in transitions from normal to ictal states.
Reviewer #1: It should be emphasized that new was obtained by analyzing the signals of the existing Wendling model. In general, the work is done and described at a good level, the results are presented clearly and in sufficient detail.

Response:
We thank the reviewer for the comments. Relevant issues were pointed out and we believe that the quality of the work has significantly improved by addressing these issues.

Reviewer #2:
The Ms describes application of neural mass model to study stimulation-based assessment of system closeness to seizure. The stimulation-based biomarkers of approaching transitions are based on a critical slowing down signal measures and are shown be able to identify proximity of the ictal state. The study is original and offers useful, conceptual insight into the problem of seizure prediction using active paradigms. The authors are familiar with the literature relevant to the problem and describe their work in this context. In general it is very good work that deserves a publication. I have only minor comments to improve the Ms.

Response:
The reviewer has demonstrated an excellent understanding of the work and we greatly appreciate the comments. Relevant and accurate remarks were presented on how to improve it, in addition to pointing out misplaced and missing references and typing mistakes that were overlooked by us.
The authors use 2 model settings -with unilateral and bilateral stimulation. The unilateral stimulation is delivered only to the side with the "normal" set of parameters. It seems counterintuitive. The most direct test would be to selectively stimulate the population with "pathological" set of parameters to see if the probing stimulus can identify the on-going process of ictogenesis. Such test is not performed. Please explain your rationale of probing "normal" network rather than "epileptic" one in the unilateral setting.

Response:
The reviewer raises an excellent point, which we had extensively discussed before deciding to use the listed model settings. The rationale for unilateral stimulation of the "normal" population is that we were looking for the "unexpected" result. That is, we expected the on-going process of ictogenesis from population 1 to be highlighted by direct probing (and this is indeed the case, as shown in the additional figure), but unsure if indirect (applied to the "normal" population) would have any predictive value -and this seems to be the case for increasing excitability, but not slow inhibition. Both strategies are interesting to evaluate in this model, as they might give insights on which conditions (in terms of stimulation target and recording site in relation to the epileptogenic zone) can an active probing approach provide predictive value. Furthermore, this may tell us something if and how this strategy can be adapted with local and remote closed-loop seizure suppression devices. Thus, we have included figures for unilateral stimulation of the ictogenic population as Supporting information.
We would also like to highlight that other model settings and stimulation parameters can be tested straightforwardly with the code available at https://github.com/vrcarva/WNMM_probing.

(Section 2.3, lines 208,209)
Unilateral probing of the ictogenic population 1 is evaluated in additional model settings, included in S1 and S2 Figs.

(Section 3.1, lines 321-324)
However, if unilateral stimulation is applied to the ictogenic population, as shown in S1 and S2 Figs 2. Figure 4, top row, shows that Variance, Lag-1 AC and to some extend Kurtosis measured in Population 1 are sensitive to B parameter change (I-B setting). In the text it is described that feature changes indicating change of B were not found except AC(1). Please clarify.
Response: This is now corrected both in the results and the discussion sections, indicating that variance and kurtosis have slight trends as B 1 is decreased.  Figure 2 is a bit confusing, partly because of panel annotations a.1, a.2, b.1 etc. I suggest to mark the panels with headings e.g. "Population 1", "Population 2", "Active", "Passive" etc. It will help to quickly get information from the figure without going through the (complex) legend. Also the role of enlarged signals is not clear. What is the reader supposed to learn from them? I suggest to simplify the figure by removing them or at least making them smaller.

Response:
We thank the reviewer for the suggestion. Indeed, marking panels with the suggested headings makes the figure much clearer, so we have modified it.
The enlarged figures should convey the shape and scale of the different types of activity (so we selected 2-3 segments) and response to stimuli of the simulated signals (with C showing the gradual amplitude and shape changes of the response). But they really take too much space, so we scaled these down on the vertical axis.