The FAST Graph: A Novel Framework for the Anatomically-Guided Visualization and Analysis of Cortico-Cortical Evoked Potentials

Abstract

Background

Intracerebral electroencephalography (iEEG) using stereoelectroencephalography (SEEG) methodology for epilepsy surgery gives rise to complex data sets. The neurophysiological data obtained during the in-patient period includes categorization of the evoked potentials resulting from direct electrical cortical stimulation such as cortico-cortical evoked potentials (CCEPs). These potentials are recorded by hundreds of contacts, making these waveforms difficult to quickly interpret over such high-density arrays that are organized in three dimensional fashion.

New Method

The challenge in analyzing CCEPs data arises not just from the density of the array, but also from the stimulation of a number of different intracerebral sites. A systematic methodology for visualization and analysis of these evoked data is lacking. We describe the process of incorporating anatomical information into the visualizations, which are then compared to more traditional plotting techniques to highlight the usefulness of the new framework.

Results

We describe here an innovative framework for sorting, registering, labeling, ordering, and quantifying the functional CCEPs data, using the anatomical labelling of the brain, to provide an informative visualization and summary statistics which we call the “FAST graph” (Functional-Anatomical STacked area graphs). The fast graph analysis is used to depict the significant CCEPs responses in patient with focal epilepsy.

Conclusions

The novel plotting approach shown here allows us to visualize high-density stimulation data in a single summary plot for subsequent detailed analyses. Improving the visual presentation of complex data sets aides in enhancing the clinical utility of the data.

1. Introduction

Patient evaluation using SEEG analysis involves either unilateral or bilateral intracerebral depth electrode implantation of a varying number of electrodes. The placement of these electrodes is determined by complex anatomical-electroclinical correlation analyses, which generate hypotheses of brain regions involved in the epileptic network in a given patient. Each electrode consists of up to 16 individual contacts which are capable of both recording data and applying stimulation. A review of the 110 consecutive patients who underwent both SEEG and CCEPs testing at the Cleveland Clinic in 2016 and 2017 reveals that 66.4% of patients had a unilateral implantation, 60.3% of which were left hemisphere implantations. The average number of electrodes implanted was 14, with an average of 144 contacts located within cortex.

The CCEPs technique evaluates the cortical evoked potentials which result from 1 Hz direct electrical stimulation applied to a pair of intracranial contacts on an electrode. A measurable evoked signal can be seen in both distant and nearby cortical regions. The CCEPs methodology has been previously described [1], [2]. In brief, a bipolar pair of electrodes are selected for stimulation with an alternating monophasic 0.2 ms pulse. Stimulation intensity is set from 1 to 8 mA using a Grass S88 stimulator (Astro-Med, Inc, RI). The resulting evoked potentials are recorded from the rest of the electrodes using a Nihon Khoden EEG system (EEG 1000, Nihon Khoden, Japan). Bandpass filters are set to 1-300 Hz with a sampling rate of 1000 Hz. The average evoked potential is computed from 30-60 trials. The number of electrode pairs stimulated per patient varied from 3 to 36 (mean 15.6). For recorded evoked potentials, a referential (extracranial) montage of the averaged response seen at each contact can contain upwards of 200 channels.

The CCEPs data are typically displayed as individual waveforms recorded in either a referential or bipolar montage. The multitude of individual responses makes analysis across multiple different stimulation pairs a complex task. A more refined and simple methodology would allow for better visualization of CCEPs over multiple stimulation pairs. With this review in mind, we describe a framework for organizing the data in a better summary presentation, the result of which we call a “Functional-Anatomical STacked area graph” (FAST graph).

Inspiration for this method of visualization is drawn largely from Byron & Wattenberg [3], In addressing the subject of legibility, they describe how “the main idea behind a stacked graph follows Tufte’s macro/micro principle [4]: the twin goals are to show many individual time series, while also conveying their sum.” This principle is particularly relevant in the case of CCEPs data, where the individual responses observed at a large number of locations are of interest, as well as the total response to stimulation at a given location. More specifically, adaptation of their stacked graph methods can be used to highlight clusters of contacts which fall within regions eliciting relatively large responses, as compared to other areas.

2. Method

The steps involved in processing the CCEPs data in order to generate a visualization summarizing the results are shown in Figure 1. Anatomical information is combined with the stimulation recordings in order to automatically determine the location of each depth electrode. This information is then used to visualize the observed responses to stimulation in a summary figure we call the FAST graph. In the sections below, we describe the details in many of these steps.

1.

Flow charts showing the steps involved in the CCEPs visualization method. Flow chart (a) shows the steps taken with the anatomical data, as well as the signal processing performed on the raw CCEPs recordings. This information is then combined to create the FAST graph (gray box). Flow chart (b) is an expansion of this step, detailing first how a single stimulation is visualized, and then how the results of multiple stimulations are combined to create the final display.

2.1. Anatomical Labelling of CCEPs

Following the planning and implantation of invasive monitoring electrodes, the location of each contact is reviewed to assign its anatomical region. SEEG contact labeling is a non-trivial and time-consuming process to complete manually. We perform the labeling process in a semi-automated manner using Brainstorm [5] to automatically assign contact labels given a segmented cortex and a list of known contact coordinate locations. The user may perform the MRI to CT registration within Brainstorm and then manually assign the electrode locations by marking the tip and skull entry in each case. The MRI of the patient is analyzed using BrainSuite [6] to produce a model of the brain which is segmented into labels as defined by the USCBrain atlas, a high-resolution single-subject atlas, constructed using both anatomical and functional information to guide the parcellation of the cortex [7]. Gyral regions within the atlas are subdivided based on resting fMRI data to create functional subparcellations [7]. Additional delineation of sulcal banks and gyral crowns in the atlas provides greater flexibility to define regions of interest.

The Brainstorm process then combines the Brainsuite segmentation and labeling with known contact locations to automatically assign a region label to each contact within the channel file. Validation of the SEEG contact labeling employed here can be found in [8].

2.2. CCEPs Data Processing

Processing of the CCEPs data was performed using Brainstorm. The raw CCEPs epochs are imported, bad epochs are removed. Each individual epoch is 1 second long with a 100ms pre-stimulation period and a 900ms post-stimulation period. The duration of each epoch is tested to determine if the recording was ended prematurely, which could result in an incomplete epoch.The remaining set is averaged to create a robust evoked response to stimulation at each contact. Given the structured nature of CCEPs testing, these steps are conveniently automated using the “process” functionality of Brainstorm [5]. We then examine and compare the response across the array for all stimulation pairs. The visualizations appearing below are all generated using custom Brainstorm processes, which provide the user with a pre-defined set of options that can be used to manipulate the data for optimized viewing.

2.3. Visualizations

An example method of displaying CCEPs can be seen in Figure 2a, which shows individual displays of the averaged response across all contacts for a stimulation at 8 mA applied to the superior temporal gyrus (S’ 13-S’ 14). Figure 2b shows all channels arranged vertically in a column, and Figure 2c shows the same channels superimposed on a single horizontal axis, at three times magnification of the scale. Comparison of the relative observed response across multiple stimulations is cumbersome for all three visualizations, and the locations of the contacts are not immediately obvious. We note that while relatively little response is observed on many of the contacts, a noticeable response can be seen on other contacts within the superior temporal gyrus (I’4-I’8, U’6-U’10), as well as on the contacts of the T and T’ electrodes located within the right and left transverse temporal gyrus.

2.

Various displays of averaged CCEPs data. (a) Individual plots of the averaged response recorded at each contact (truncated at 600ms for the display) resulting from stimulation of the S’ 13-S’ 14 contact pair (denoted by STIM-STIM). (b) The same data is shown with even vertical spacing, and (c) superimposed on the same vertical axis. (d) Shows just the contacts from the T electrode, and (e) an unsorted example of the area plot of the same T electrode data, where the responses are layered to indicate the total response observed on the electrode as it changes over time.

Figure 2d shows a closer look at the observed response on the T electrode, where the number identifies each individual contact. If we rectify each waveform, then a stacked “area plot” layers the individual responses and colors the interval between waveforms, as shown in Figure 2e, generated by the ‘area’ plot function in MATLAB (the Mathworks, Inc., Natick, MA, USA). The function assigns different colors for adjacent contacts to make differentiation between them easier, and the T5-T8 contacts stand out more clearly as showing the largest responses. The height of the overall graph represents the total summed response across all contacts, also known as the “L₁ norm.”

In the case of the stacked area graph shown in Figure 2e, the cycles of colors are repeated for each electrode, with adjacent contacts colored so as to make them easily distinguishable from adjacent contacts, but are otherwise arbitrarily assigned. When displaying all electrodes on a single plot, the color scheme must be modified such that contacts represented by the same color relate to each other in a meaningful way, and conversely that contacts with no relation are distinguishable from each other. With this in mind, additional information can be encoded into the figure by using color to represent the location of each contact within the patient’s brain. This process is described in detail below.

2.4. FAST Graphs

The steps involved in generating FAST graphs are summarized in Figure 1, and each step is described in detail in the following paragraphs. Briefly, the channels are separated by hemisphere above and below the x-axis, sorted based on their temporal energy, then rectified and stacked in an area graph. The area under the curve for each contact is colored to represent its anatomical region as indicated by the legend in the figure. Each subplot represents the average of a single stimulation pair, and we display the entire stimulation protocol in the same figure to highlight differences in relative responses across stimulation sites.

The horizontal axis serves as a natural means by which to distinguish between contacts located in different hemispheres of the brain. We therefore employ a stacked graph on either side of the x-axis, such that the responses of contacts in the left side of the brain are plotted above the horizontal axis, and those in the right side are plotted inverted below it.

In Figure 2, the waveforms are ordered simply by contact number; however, additional information can be represented in the graph by reordering the waveforms according to a user-defined set of criteria. In the FAST graph, we choose to sort the CCEPs based on the average standard deviation of each waveform over pre-defined intervals, such as the “early” (10 – 60 ms) or “late” (60 – 250 ms) responses after the electrical stimulus. Arranged in ascending order, the contacts with the smallest standard deviation appear closest to the horizontal axis. Alternatively, we could sort based on the absolute max response seen at each contact. In applying such a sorting method, it proves beneficial to limit the time range over which the data are sorted, for example the time period immediately after stimulation can be excluded in order to avoid any corruption by stimulation artifact. Implementation of these techniques as a process within Brainstorm presents the user with a predefined selection of sorting methods, as well as the option to customize the window of time over which the data should be sorted. The result of sorting the responses in this manner is that the regions of the brain contributing most to a given response become much more easily identifiable than in a montage ordered merely by the contact name.

Each contact has an associated spatial segmentation label (e.g., post-central gyrus), and we have clustered these labels into the six broad regions: pre-frontal, frontal, temporal, central, parietal, and occipital. By assigning each region a color, the user is provided with a better general indication of which regions elicit the largest response to stimulation. A legend showing the regions and the color associated with each region appears alongside the graph. A version of the legend can be seen in Figure 3, along with a graph showing the results of stimulation of the left superior temporal gyrus for a particular patient. A relatively larger response is observed in many of the contacts located in the left temporal region.

3.

(a) The FAST graph of the stimulations performed on a particular subject at 8 mA. The response at each contact is colored according to its’ location as shown in the inset. This shows a large response in the temporal lobe when the S’ 13-S’ 14 contacts (located in the left superior temporal gyrus) were stimulated. (b) Plots showing periods of statistically significant response for three of the stimulations, calculated using 10,000 sign permutations and corrected for multiple comparisons using the false discovery rate of 0.05. The true measured L₁ norm is shown in black, with periods of significance indicated in bold. The grey lines represent the L₁ norm for all of each of the sign permutations. The R1-R2 stimulation shows no significant response, the Q’1-Q’2 stimulation shows an early period of significant response. The S’13-S’14 stimulation shows a much larger and sustained significant response.

With the FAST graph designed and generated for a single simulation contact pair, we can then order many FAST graphs into a single figure to get an overview of spatial or temporal activity. For CCEPs data, we sort the FAST graph subplots by the stimulation location, ordered anterior to poster on the left side, followed by anterior to posterior on the right side, as shown in Figure 3. In order to make direct comparisons easier, the stacked graphs are displayed on a uniform scale. The resulting montage of individual FAST graphs into a single overall FAST graph figure gives a summary indication of which areas produce a more robust response than others. A complete display of the CCEPs stimulations performed on a particular subject at 8 mA can be seen in Figure 3.

2.5. Variations

We consider an implantation in which the vast majority of contacts are located within the temporal lobe. Coloring the responses by just the temporal region yields little additional information, since only a single region color dominates the FAST graph. In the case of localized implantations, a more targeted display of the contact locations can convey meaningful information. We optionally refine the FAST graph by assigning each cortical label a unique color, to visualize where exactly within a region the largest responses are being recorded. Figure 4 shows a FAST graph where the temporal responses are colored according to each segmentation label within the temporal lobe. In the case of an implantation where certain regions are more densely sampled than other regions, we can account for this by normalizing the responses according to the number of contacts located in each region. In addition to this, we can also separate out the responses into local (less than 2 cm from the stimulation site) and distant (greater than 2 cm from the stimulation site). Examples of this are shown later in Figure 6.

4.

(a) Here the contact locations are not colored by coarse regions but rather by the color of the individual segmentation labels to provide more detailed information. (b) A FAST graph showing the responses to stimulation for an implantation which is confined to the temporal lobe. The yellow in the A’ and mesial E’ stimulations indicate the largest response to be coming from the parahippocampal gyrus.

6.

Ictal EEG in SEEG evaluation. Ictal SEEG changes started in the left medial superior frontal gyrus (contacts L’1-5), preceding clinical onset by 4.5-9.5 seconds.

2.6. Statistical Analyses

The FAST graph provides a summary of the relative strength of the response to stimulation for each pair of contacts. We observe that each stimulation appears to elicit a significant response compared to the pre-stimulation interval; however, some stimulations evoke much greater responses than others. The question naturally arises as to whether or not these responses are statistically significant.

We note that the envelope of the stacked graphs represents the sum of the absolute data and is therefore the L₁ norm response of the array, as a function of time. In order to assess statistically the significance of the observed L₁ norm, we can apply a “sign permutation test” following the procedure of [9]. Each stimulation consists of 30 pulses, one per second, which we refer to as the 30 trials. We randomly multiply each of the 30 trials by +1 or −1, then form a new average evoked response. Our null hypothesis is that no response exists, and therefore changing the sign of random noise should not statistically effect the average. Repeating this sign permutation procedure ten thousand times enables us to develop a data driven distribution of the L₁ norm for the array under the null hypothesis. We can then determine significant responses as those locations on the time curve that lie in the tail of the null distribution. We use the false discovery rate procedure [10]¹ to control for multiple comparisons with a FDR of 0.05.

Figure 3(b) shows significance results for three of the stimulation pairs. The gray portion of the plot shows the overlay of 10,000 sign permuted L₁ norms, showing the data-driven distribution under the null hypothesis. The thin black line shows the actual Li norm of the array under question, overlaid on the distribution. An asterisk is assigned for each time point in which the L₁ norm is significantly different from zero. The left plot indicates that no significant response was evoked with stimulation of the right postcentral gyrus at contacts R1-R2. The center plot shows a short period of relatively minor, yet nonetheless significant response to stimulation of the left pars opercularis at contacts Q’1-Q’2. The right plot shows a much larger response with an extended period of significance resulting from stimulation of the left superior temporal gyrus at contacts S’13-S’14.

The sign permutation test reveals if the L₁ norm of the response is significant. A separate but related question is whether or not a response to one stimulation pair is significantly different from the response to a different stimulation. In order to assess relative significance between two separate stimulations, we employ a permutation test to determine the difference of means. An example of this is shown later in the Applications section, where Figure 5 includes a plot comparing the L₁ norms across the stimulation of the L’3-L’4 and X’1-X’2 pairs. Our null hypothesis is that there is no difference in their L₁ norms, which we test by permuting trials between the two stimulation sites before computing the mean response and L₁ norm. As before, we use FDR to control for multiple comparisons.

5.

Summary of non-invasive and intracranial evaluation. (A) The MEG showed a cluster of interictal dipoles in the left frontal pole (90%) and right frontal pole (10%). The subtraction ictal and inter-ictal SPECT co-registered to MRI (SISCOM) presented ictal hyperperfusion in left anterior medial frontal lobe. The 18-F-fluorodeoxyglucose positron emission tomography (FDG-PET) imaging showed hypometabolism in the left anterior medial frontal lobe. The VBM study revealed an abnormality suggesting a possible lesion in the left anterior medial frontal lobe. (B) Configuration of the SEEG implantation. (C) Continuous interictal epileptiform discharges were recorded in the left medial superior frontal gyrus (contacts L’2-L’4). (D) Ictal SEEG changes started in the left medial superior frontal gyrus (contacts L’1-L’5).

3. Applications

We now consider two select examples and discuss observations occurring from the FAST graph as they relate to the clinical data.

3.1. Case 1

The patient presents as a 17-year-old right handed male with seizure onset at 9 years. He has a prior diagnosis of left frontal epilepsy and had failed eight prior anti-seizure medication trials as well as vagus nerve stimulation. His initial seizures were characterized as staring episodes and periods of inattention lasting 10-15 seconds. He described an aura of “a cold chill in my back” or “an empty feeling in my head.” The frequency of his seizures was about 20 per day, almost every hour during sleep, and always focal with impaired consciousness. The ictal EEGs were localized to the left frontal region or were non-localizable. MRI was non-lesional but voxel based morphometric (VBM) analysis showed a possible lesion in the left anterior medial frontal lobe. PET showed left anterior mesial frontal hypometabolism. Ictal SPECT hyperperfusion was seen in the left anterior mesial frontal area. MEG indicated an interictal tight dipole cluster in the left superior frontal gyrus. The data converged to support a left frontal epilepsy possibly arising in the left prefrontal, frontopolar or mesial frontal/anterior cingulate regions. SEEG implantation was recommended to explore this hypothesis, including exploration of the anterior insula, operculum, and parietal areas which are well connected to prefrontal regions. The SEEG evaluation showed a very restricted epileptogenic zone, localized in the anterior-mesial aspect of the left superior frontal gyrus (contacts L’2 through L’4). The ictal patterns consisted of rhythmic spike and wave activity at electrodes L’2 through L’4 followed by fast activity in the same electrodes and L’8 to L’ 10 as well as X’1-X’2 within 0.5 to 2 seconds from ictal onset. Direct cortical stimulation induced typical seizures via stimulation of L’3-L’4 contacts. Summaries of the clinical findings are shown in Figure 5 and Figure 6. The patient underwent a laser ablation targeting only the mesial trajectory of the L’ electrode. The laser ablated lesion diameter measured 15 mm. The patient is completely free of both seizures and auras for the last four years. In this case, the mesial contacts of the L’ electrode (L’3-L’4) represents an exquisitely small epileptogenic zone in this patient.

A 3D reconstruction of the implanted electrodes obtained using Brainstorm described above can be seen in Figure 7a–b. Figure 7a shows the labels represented by each color within the montage (SEEG electrodes colored black to appear more visible). Figure 7b provides an alternative display which colors the spheres representing the contacts according to their label. This view is particularly useful in the case of contacts on the boundary of regions, showing exactly what portion of the contact lies in each region. Note that although all electrodes have left hemisphere notation, the most mesial F’, O’ and X’ contacts pass through into the contralateral hemisphere.

7.

(a) A frontal implantation; (b) the same implantation, but now each contact is colored by its spatial label; (c) FAST graph showing the largest response is to the stimulation of the L’3-L’4 contacts, and second largest response to the stimulation of the L’3-L’4 contacts, and second largest response to the stimulation of the X’1-X’2 contacts. (d) A comparison of the L₁ norm (envelopes) of these two stimulations, where asterisks denote instances when then two responses are significantly different from each other, as determined by permuting the two sets 1,000 times and applying the false discovery rate of 0.05.

CCEPs stimulation was performed in this patient at five stimulation pairs chosen based on their involvement in the ictal and interictal activity, the results of which are also shown in Figure 7. With the SEEG implantation confined to frontal/pre-frontal areas, visualizing the data with individual label colors provides finer detail than the regional colors. A notably large response is seen when stimulating the L’3-L’4 pair, located in the anterior superior frontal gyrus.

Since the superior frontal gyrus is more densely sampled than other regions, we review the responses normalized according to the number of contacts located in each region, and separated out into local (less than 2 cm from the stimulation site) and distant (greater than 2 cm from the stimulation site) responses. Figure 8 illustrates the difference in connectivity patterns from near versus far regions in relationship to the stimulation site. These differences may be meaningful as it relates to the early spread zone of the ictal onset zone and is consistent with the ictal propagation networks in this case example.

8.

Variations of the FAST graph. In both sets the responses are normalized according to the number of contacts within each region. In graph (a) only “distant” responses are shown (further than 20mm from the stimulation), graph (b) only “local” responses are shown (less than 20mm from the stimulation). Noting the change of scale in (b), we observe the near networks in (b) implicate the sublobar area L’3-L’4 ablated by laser therapy.

Here we note that as shown in Figure7, the CCEPs stimulations reveal a larger response to L’3-L’4 stimulation with the resultant CCEPs response from the left anterior superior frontal gyrus indicating the largest contribution to the response. The next most prominent response is seen with stimulation of X’1-X’2 (left mid-cingulate gyrus), giving rise to the robust CCEPs response from the left inferior frontal gyrus. As discussed above in the section on statistical analyses, we performed a permutation analysis of the two sets of data, and as Figure 7 reveals, we found a statistically significantly larger response to L’3-L’4 stimulation, compared to the X’1-X’2 stimulation. This larger response is consistent with L’3-L’4 being in the epileptogenic zone and with X’1-X’2 being within the near propagation network. This finding is congruent with the rest of the patient’s neurophysiologic and imaging data.

3.2. Case 2

A 39-year-old ambidextrous female with medically refractory seizures presents with seizures consisting of a loss of awareness. Scalp video-EEG monitoring showed interictal epileptiform discharges arising from both left and right anterior temporal regions. The typical clinical seizures showed ictal EEG changes that were classified as left frontotemporal, right frontotemporal, or non-localizable. The MEG showed a tight dipole cluster in the left anterior temporal region as well as a single dipole in the right temporal region. Both 3T brain MRI and VBM were normal. FDG-PET showed hypometabolism in both temporal lobes, more pronounced in the left temporal lobe. SISCOM illustrated ictal hyperperfusion in the left hippocampus. Noninvasive evaluations led us to perform the SEEG evaluation based on the hypotheses of epileptogenic zone in the left (prominent side) and right temporo-perisylvian regions. Figure 3b represents the 3D-MRI with SEEG electrodes. Interictal epileptiform discharges were recorded in the left mesial temporal (40%, mesial part of electrodes B’, A’, and E’), the right mesial temporal (40%, mesial part of electrodes B and A), and the left anterior lateral temporal regions (20%, lateral part of electrode J’). Ictal SEEG changes started in the left mesial temporal region (mesial part of electrodes B’ and A’), which showed early spread to the contralateral right mesial temporal region. The SEEG evaluation revealed the epileptogenic zone in the left mesial temporal region (head of hippocampus and amygdala). Summaries of the clinical findings are shown in Figure 9 and Figure 10.

9.

Summary of non-invasive and intracranial evaluation. (A) The MEG showed the tight cluster of dipoles over the left anterior temporal lobe and a single dipole in the right temporal lobe. SISCOM presented ictal hyperperfusion in the left hippocampus. FDG-PET showed hypometabolism bilateral temporal lobes more pronounced in the left temporal lobe. (B) 3D-MRI with SEEG electrodes in the left and right temporo-perisylvian regions. (C) Interictal epileptiform discharges were recorded in the left mesial temporal (40%, mesial part of electrodes B’, A’, and E’), the right mesial temporal (40%, mesial part of electrodes B and A), and the left anterior lateral temporal regions (20%, lateral part of electrode J’). (D) Ictal SEEG changes started in the left mesial temporal region (mesial part of electrodes B’ and A’), which showed early spread to the contralateral mesial temporal region.Ictal EEG in SEEG evaluation.

10.

Ictal EEG in SEEG evaluation. Ictal SEEG changes started in the left mesial temporal region (mesial part of electrodes B’ and A’), which showed early spread to the contralateral mesial temporal region in 2-3 seconds.

In this case a much larger number of electrode pairs (not limited to interictal or ictal regions) were stimulated during the CCEPs testing as shown in Figure 11. This highlights the benefit of summarizing CCEPs responses from a large number of stimulation pairs. Clearly visualized in the figure are notable large responses to stimulation of the B’3-B’4 contact pair, followed by the responses to stimulation of the B3-B4 contacts. These regions reflect the predominant regions of interictal epileptiform activity seen in the SEEG evaluation left and right amygdala and hippocampus. The ictal SEEG findings arising from the left amygdala and hippocampus correlate well with the largest CCEPs response seen in the graph. This patient is scheduled to undergo a left temporal resection in the near future.

11.

FAST graph of 4mA CCEPs stimulations. Notable responses to stimulations are seen for the B’3-B’4 and B3-B4 stimulation pairs.

4. Discussion

We have developed a new methodology for summarizing high density CCEPs data in a concise and principled graph, using the spatial information of the contacts. The waveforms are first sorted based on the RMS or peak of the waveform, then the data are grouped by hemisphere before coloring in an area graph. The colors of the area graph are assigned by the spatial region or label of the contact. Each area graph is then arranged within the overall figure based on the site of stimulation. The resultant “FAST graph” thus gives us a concise summary of the effects of stimulation across all stimulation sites and across all recording contacts. We further demonstrated that the envelopes of the graphs can then be statistically treated to reveal significant responses and significant differences among the waveforms.

Single pulse electrical stimulation has been used to probe cortical excitability in focal epilepsy. The resulting cortical responses has been divided by some authors into early and delayed. The presence of delayed responses occurring at latencies between 100ms to 1.5s appear to be correlated to the epileptogenic zone [12]. Others have focused on accentuation of amplitudes of the initial components of CCEPs which may be reflective of the increase excitability of the epileptogenic zone [13].

This FAST graph approach could be applied to a variety of neurophysiologic data such as CCEPs, ECoG, and MEG. The ability to concisely review complex data such as CCEPs helps potentially in its application across data sets. CCEPs results involve multiple evoked potentials across large arrays following one-paired stimulation. The need to compare the CCEPs results across multiple stimulation pairs quickly benefits the further analysis of the result. As discussed in Figure 4, the user is provided the ability to display a desired subset of the implanted electrodes by including/excluding contacts from particular regions. This is important not only to allow the user to focus on individual regions, but also due to the varied nature of implantations.

The construction of the CCEPs directed networks could also be visualized in terms of distance. The short-range connection can be visualized in limiting responses seen within various Euclidean distance from the midpoint of the stimulation pair as visualized in Figure 6 which separates out the responses within 2 cm. This analysis allows visualization of the effect of distance on CCEPs [14].

The illustration of multiple stimulation pairs in a single individual allows for observable associations with other data sets. In the case example, we illustrated how ictal SEEG data analysis agrees well with CCEPs stimulation results, in a patient in whom a small lesion created by laser ablation resulted in a seizure free outcome. The accentuated response seen in the CCEPs results is greatest at the ictal onset zone. Following construction of the CCEPs constrained by distance, a significant component of the accentuation is seen within 2 cm of the stimulation response. The accentuated CCEPs responses near the ictal onset zones reflect increased excitability of the cortex associated with the epileptogenic zone [15]. The ability to survey these large data sets quickly enhances the clinical utility of CCEPs.

5. Conclusion

The novel plotting approach demonstrated here allows us to combine spatial and temporal information in a single summary plot, over many stimulation points and with high-density arrays. We exploit the advanced and automated 3D segmentation analysis of MRI software such as Brainsuite to identify, order, and color the contacts. The waveforms are ordered within the subplots by user preference, then the area plot is colored by these regions. The resulting FAST graph provides a powerful summary that draws attention to relevant data that needs further detailed analyses. The improved visual presentation of complex data sets could enhance the clinical utility of other neurophysiologic data, particularly when comparing across modalities such as ictal electrocorticography and CCEPs in patients undergoing SEEG.

Highlights.

• Functional-Anatomical Stacked area graphs for cortico-cortical evoked potentials

• The FAST graph summarizes entire CCEPs stimulation protocols in a single figure

• Combining spatial and temporal information improves visual presentation

• Novel measure of connectivity showing significant interactions of brain regions