If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Accurate source imaging based on high resolution scalp electroencephalography and individualized finite difference head models in epilepsy pre-surgical workup
Department of Neurosurgery, Huashan Hospital, Fudan University, Shanghai 200040, ChinaDepartment of Neurosurgery, Jing’an branch of Huashan Hospital, Fudan University, Shanghai 200040, China
Individualized finite difference model method is feasible in clinical work.
•
EEG source imaging technique localizes epileptogenic zone accurately.
•
EEG source imaging technique contributes to invasive evaluation plans for epilepsy.
Abstract
Purpose
High-density electroencephalographic source imaging (HD-ESI) has emerged as a useful tool for pre-surgical epilepsy workup. However, it is not routinely used in clinical evaluations due to several factors, one of which is the challenge associated with creating anatomically accurate head models. Reasonable solutions now exist and the present study aims to evaluate the use of these highly resolved individual head models in pre-surgical epilepsy evaluation.
Methods
Nine patients with intractable epilepsy who were candidates for resective epilepsy surgeries participated in the study. For each patient, 256-channel electroencephalography data were acquired along with individual structural MRI data that was used to construct individual finite difference models (iFDM). Accuracy of HD-ESI based on iFDM (HD-ESI-iFDM) was evaluated using multiple criteria, including concordance with intracranial electroencephalography (icEEG) and location of surgical resection. Performance of HD-ESI-iFDM was also compared against MRI and positron emission tomography (PET) results.
Results
In all but one patient resective surgeries resulted in seizure-free outcome. Source locations derived from HD-ESI-iFDM demonstrated concordance with surgical resection and with icEEG data, when available. The HD-ESI-iFDM also contributed to the planning of intracranial electrodes implantation. Compared to MRI or PET, HD-ESI-iFDM provided more accurate localization of the epileptogenic zone.
Conclusion
When acquired with high-density sensor arrays and source imaging is performed with anatomically accurate head models, electroencephalography can contribute meaningfully to epilepsy pre-surgical workup for localization of the epileptogenic zone. Now that both high-density electroencephalography and individualized FDM models can be routinely obtained, it can be incorporated as part of clinical practice.
Electroencephalography (EEG) provides definitive epilepsy diagnosis. However, within the context of pre-surgical workup, its value has not been fully realized because very sparse sensor arrays (commonly 19-channels) are often used. It has been shown that for accurate estimates of inter-ictal spikes sources, denser channel counts are required [
]. Moreover, to fully take advantage of the higher resolution scalp data for localizing cortical sources of seizures, source estimates must be performed analytically, as inferences of sources from scalp data are inaccurate, due to volume conduction and overlap of multiple sources at any given point in time. With adequate sampling of the EEG field provided by dense-sensor arrays and accurate head models of current flow, studies show have shown that high-density electroencephalographic source imaging (HD-ESI) can accurately localize the epileptogenic zone in a majority of cases, with sensitivities ranging between 84% and 97% [
]. However, these prior studies used different resolution head models, such as spherical and atlas-based finite difference method (FDM) models for ESI. Head model methods vary with respect to how accurately they can describe current flow. Because ESI is critically dependent on accuracy of the head model used, accuracy for estimating the epileptogenic zone will vary according to fidelity of the method used to generate these models.
Individualized (i.e., based on the anatomy of each patient) finite element method (FEM) and FDM models provide the highest fidelity but are often not employed routinely because they are difficult to construct or computationally demanding. FEM models are difficult to construct because of strict topological requirements of the mesh generation process, and inaccuracies affect the ability to calculate current flow. FDM models, on the other hand, can be constructed directly from the voxel structure of the magnetic resonance imaging (MRI) data, and thus avoids difficulties associated with generation of the mesh for each compartment of the head. The computational demand of the FDM approach is prodigious, limiting its routine use. However, given recent developments in computational power and affordability, FDM models can be computed quickly, requiring less than one hour to construct. In this study, we examine the performance of HD-ESI based on individualized FDM models (HD-ESI-iFDM) to identify the epileptogenic zone.
2. Methods
Nine consecutive patients (males = 6, females = 3, average age = 23.8 years) who consented to HD-ESI-iFDM evaluation, resective surgery, and long-term follow-up were enrolled in the present study (see Table 1). In addition to the HD-ESI evaluation, patients also had conventional pre-surgical workup, which included evaluation using 16-channel scalp video EEG, semiology assessment, Fluoro-2-deoxy-d-glucose positron emission tomography (FDG-PET) imaging and radiological review of MRI data for potential structural abnormalities.
Table 1Pre-surgical workup results of multiple tools, surgical resection and outcome summary.
MRI
FDG-PET
HD-ESI-aFDM
HD-ESI-iFDM
Ictal icEEG
Interictal icEEG
Resective area
Time of FU (mos)
Case 1
Negative
L Ant Temp
L Ant Temp
L Ant Temp
–
–
L Ant Temp
55
Case 2
L Med Temp
L Temp (broad)
L Med Temp
L Med Temp
–
–
L med Temp
56
Case 3
Negative
L Ant Temp
L Ant Temp
L Ant Temp
–
–
L Ant Temp
54
Case 4
L Post Temp
L Temp (broad)
L Post Temp
L Post Temp
–
–
L Post Temp
54
Case 5
L Med Temp
L Med temp
L Temp Pole
L Med Temp
–
–
L Ant Temp
54
Case 6
R Temp. Pari
R Temp Pari (broad)
R Post Temp
R Post Temp
–
–
R Temp Pari
54
Case 7
R Ant Temp
R Temp Pari (broad)
R Post Temp (basal)
R Post Temp (near TPO)
R Post Temp (near TPO)
R Post Temp (near TPO)
R Ant Temp + R Post Temp (near TPO)
49
Case 8
Negative
R Fron
R Sup Temp Inf Fron
R central lobe
R central lobe
R central lobe
R central lobe
28
Case 9
R Lat Fron Pari
R Lat Fron Pari
R Med Fron
R Med Fron
R Lat Fron Pari
R Lat Fron Pari
R Lat Fron Pari
50
Abbreviations: Ant = anterior; Med = medial; Lat = lateral; Fron = frontal; Temp = temporal; Post = posterior; Pari = parietal; R = right; L = left; Sup = superior; Inf = inferior; TPO = temporal-parietal-occipital region; FU = follow-up; mos: months.
A 256-channel high-density EEG (HD-EEG) system (Philips Neuro, Eugene, Oregon, US) was used to record the scalp potential field. Each EEG sensor net was applied using the nasion, bilateral pre-auricular locations, and Cz position as landmarks to ensure standardized placement across patients. The tension structure of the net ensured even distribution of the remaining sensors on the head and at similar locations across patients. The EEG was recorded with a DC amplifier and sampled at 500 or 1000 s/s. Photographs of patients with the sensor net were acquired for use to construct the FDM head models.
Dominant interictal spike types from the HD-EEG data were identified by two experienced EEG experts. Topographic maps were used to examine spatial distribution, and spikes were grouped according to spatial similarity prior to spike averaging. The time-point around 50% ascending phase of the dominant, averaged spike type for each patient was chosen for HD-ESI [
In addition to use of a generic FDM atlas for HD-ESI, iFDM models were also created. To construct iFDM models of each patient, high quality axial T1-weighed MRI data were acquired using the 3D-SPGR sequence (General Electric, US) with 1 × 1 × 1 mm resolution, scanning from top of head to the chin. Tissue segmentation, with the skull estimated through use of an atlas skull registered to the individual MRI, was performed to identify the following head tissues: gray matter, white matter, cerebral spinal fluid, skull, eyeballs, air, and flesh [
]. EEG sensor positions (average positions with manual identification of nasion, preauricular landmarks, and proximal eye and ear sensors in photos of each patient wearing the sensor) were then registered to each individual’s model. Oriented current generators (orientations are based on morphology of the cortical surface) were distributed (1200 per hemisphere) throughout the cortical surface.
Once constructed, computation of the lead field (i.e., a description of current flow from the current generators throughout the head volume to the sensor positions) was performed with the following average, literature-based conductivity values: gray matter = 0.25 Siemens/meter (S/m), white matter = 0.35 S/m, CSF = 1.79 S/m, skull = 0.01 S/m, eyeball = 1.55 S/m, and flesh = 0 0.33 S/m. The workflow for construction of the iFDM models is presented in Fig. 1.
Fig. 1Construction of individualized high-resolution finite difference method head model summary.
] was used as the inverse method to generate the ESI estimates. Accuracy of HD-ESI results was then compared using multiple criterion, including the “gold standard” intracranial EEG (icEEG) and location of successful surgical resection.
3. Results
Of the nine patients, after long-term follow-up (range = 28–56 months, average = 50.4 months) eight achieved excellent surgical outcome (Engel grade I) while one (NO. 9) did not fully respond to treatment (Engel grade III). Because determination of accurate localization of epileptogenic zone can only be accomplished in cases wherein surgical resection result in the patient being seizure-free, we evaluated performance of the various pre-surgical workup tools with respect to the eight seizure-free cases.
Accuracy of HD-ESI-iFDM was first determined through evaluation of concordance with three criteria: icEEG findings (cases NO.7 and 8), resective area (MRI negative, PET positive temporal epilepsy cases NO.1 and 3), and proximity to lesions (lesional epilepsy cases NO.2, 4, 5, and 6). The results showed that HD-ESI-iFDM provided good concordance in all cases.
Next, concordance evaluations were performed using the “resected area” as the criterion for all cases, accuracy of multiple tools were: HD-ESI-iFDM 7/8 = 87.5%, HD-ESI with atlas model (HD-ESI-aFDM) 5/8 = 62.5%, FDG-PET 3/8 = 37.5%, MRI 6/8 = 75.0%.
FDG-PET results were usually broadly localizing (i.e., the localization is usually larger than the resected area), and therefore concordance could only be determined at a gross lobule or hemispheric level. Thus, for these broadly localizing cases, they were classified as failure cases relative to the resected area. MRI failed to provide localizing information in two cases because lesions could not be identified: one was diagnosed as MRI negative and the other “MRI negative” case showed only an arachnoid cyst in the contralateral hippocampal region.
HD-ESI-aFDM was not accurate in localizing case NO. 6, 7 and 8 (see Fig. 2). Furthermore, in case NO. 5, although HD-ESI-aFDM was judged as “accurate”, because it overlapped with the resected area, the results obtained from HD-ESI-iFDM was deemed as more precise due to its localization being closer to the mesial temporal lesion. HD-ESI-iFDM localized precisely in all but one case, with the exception showing diffuse sources that appeared to partially overlap the resection.
Fig. 2Case NO 7 A. interictal epileptic discharges recorded by icEEG are mainly located at electrodes NO. 7, 8, 15, 16, 22, 23, 24, 51, 52, 53; B, C. topographic view of 256 channel high-density electroencephalography showed focal epileptic discharges (blue arrows) mainly in the right mid-posterior temporal region; topographic map show energy distribution of HD-EEG with good signal to noise ratio. D. FDG-PET showed broad hypometabolism involving right temporal and parietal region; E. structural MRI indicated an anterior temporal abnormality with hyper-intensity; F. HD-ESI-aFDM showed sources located in the basal temporal region on the right side; G: co-registration of HD-ESI-iFDM and subdural icEEG electrodes showed that intracranial electrodes detecting interictal epileptic discharges (IEDs) overlap with the strongest source estimates. This area was then resected during stage two surgery and the patient achieved seizure free status. (topographic map: red = positive, blue = negative, white = zero; HD-ESI: patches with lighter colors indicate stronger source estimates; icEEG: 100uV/mm, 1Hz–35 Hz, 3 cm/s; high-density EEG: 1–70 Hz, 3 cm/s, 10 uV/mm, average reference). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 3Case NO 8 A. 3-D topographic map (red = positive, blue = negative, white = zero). B & C. HD-ESI-iFDM generated epileptogenic foci mainly around central sulcus (B: original brain; C: inflated brain showing sources in sulci and in between gyri). D. topographic view of 256 channel HD-EEG showed focal epileptic discharges (blue arrows) mainly in the right temporal parietal region. E. HD-ESI-aFDM showed sources located in the right superior temporal parietal region. F. subdural grids were planted covering mainly the central lobes. G. resection area was confirmed by icEEG results, which is concordant to the HD-ESI-iFDM estimates. (HD-ESI: patches with lighter colors indicate stronger source estimates; HD EEG: 1–70 Hz, 3 cm/s, 10uV/mm, average reference). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
HD-ESI-iFDM also contributed to the plans of icEEG electrodes implantation. In case NO. 7, icEEG was placed in the anterior temporal region because of the existence of a structural lesion. In addition, because of the HD-ESI-iFDM findings, we placed icEEG electrodes in the posterior temporal regions. This region ultimately proved to contain the epileptogenic zone. In case NO. 8 the HD-ESI results were used to guide placement of icEEG electrodes over the central lobe, because the MRI finding was not definitive (i.e., it revealed a very subtle abnormality in this region).
4. Discussion
Accuracy of HD-ESI requires a good solution to the forward problem, which is the description current propagation from cortical sources to the scalp. Prior studies using ESI are usually based on realistic forward models (FDM or FEM) constructed using an atlas MRI, because atlas models can be computed once and used on any patient. When using individual MRI data, and again because they can easily be computed, more simplistic computational (such as spherical or boundary element) models are often used [
]. Thus, it is reasonable to hypothesize that if individualized FDM models are utilized, enabling characterization of the individual’s cortical morphology and high-resolution description of volumetric current flow, HD-ESI accuracy should be improved.
The computationally demanding FDM approach has recently been improved and is now feasible at our site [
]. Using individualized FDM models, the present study provided preliminary results consistent with this hypothesis. The results showed that compared to all other evaluation approaches routinely used in pre-surgical workup, HD-ESI-iFDM was most accurate.
Results from HD-ESI obtained with individualized FDM models in the present study also enabled simultaneous spatial analysis of the structural lesion and epileptogenic zone as part of the pre-surgical workup. Results from such analysis contributed to intracranial electrodes implantation (phase I surgery) and helped with phase-II surgery planning, when electrodes are co-registered to the 3-D HD-ESI data.
Interictal EEG activities are frequently used for source imaging because they are usually discrete and can be averaged to provide a good signal-to-noise ratio, and several studies have shown, consistent with the present results, that careful analysis of interictal epileptic discharges can lead to accurate localization of the epileptogenic zone [
]. Because ictal EEG onset is generated by the epileptogenic tissue, localization of ictal onset EEG data may be even more informative. However, ictal onset is not commonly used in ESI because it is difficult to determine actual onset in the EEG time-series data. To apply HD-ESI to ictal EEG onset, methods such as joint time-frequency decomposition and functional connectivity analysis have been developed, and the initial results are promising [
Localizing value of electrical source imaging: frontal lobe, malformations of cortical development and negative MRI related epilepsies are the best candidates.
Localization accuracy of distributed inverse solutions for electric and magnetic source imaging of interictal epileptic discharges in patients with focal epilepsy.
] have been performed to demonstrate the concordance of ESI estimates to icEEG results, but only a few have used high-resolution individualized head models. Because icEEG is deemed as the “gold standard” for localizing epileptogenic zone, it will be important to carry out further validation studies comparing icEEG (either subdural or stereotactic electrodes) and HD-ESI-iFDM. To demonstrate concordance of HD-ESI-iFDM with icEEG, more spatially resolved inverse methods may be required. Linear inverse methods, such as the one used in the present study, are smeared by their very nature. However, they appear to be appropriate to the underlying generator of epileptogenic activity, such as interictal spikes [
], wherein the activity is believed to be spatially extended.
5. Limitations and future directions
The present study has several limitations that are due to the preliminary nature of the present research. First, it is well known that important current paths, such as the optic canals and foramen magnum, cannot be characterized using BEM models. These conductive paths are crucial to model accurately because sources close to the face (e.g., orbitofrontal cortex and anterior temporal lobes) and neck (e.g., basal temporal lobe) are frequently involved in many epileptic discharges. Thus, the major advantage of the FDM (and FEM) models compared to BEM models has practical implications for accuracy of source estimates for pre-surgical workup, as shown in recent publications [
]. Therefore, future studies should compare FDM, FEM and BEM models within this specific context.
Second, because we also used standard literature values for tissue conductivities, individualized conductivity estimates, should improve the accuracy of the head model, particularly if accurate conductivity can be estimated for the most resistive tissue, which is the skull. Future studies should address this important topic. Third, because of resource limitations within our lab, we had to use average EEG sensor locations in the head model, which in principle could compromise accuracy of the head model. This possibility was mitigated by use of one skilled technician during sensor net application and use of photographs to verify placement of sensors on the head model. Availability of EEG sensor localizing technologies, such as multi-camera geodesic photogrammetry system (GPS) [
] are available and can readily be used. Future studies will need to assess the influence of sensor registration accuracy on ESI accuracy.
6. Conclusions
HD-EEG data is now readily available for clinical use and coupled with the ability to construct iFDM, HD-ESI is feasible and can significantly contribute to pre-surgical workup of epilepsy surgeries, even with the limitations noted above. With improvement of modeling/source imaging methods and further validation studies, HD-ESI-iFDM may become the primary evaluation tool used for epilepsy pre-surgical planning.
Conflict of interest
None.
Study funding
This work was supported by the National Natural Science Foundation of China (NO. 81701273 and 81701250).
Acknowledgements
We thank Drs. Phan Luu and Don Tucker for their help in data processing.
References
Lantz G.
Grave de Peralta R.
Spinelli L.
Seeck M.
Michel C.M.
Epileptic source localization with high density EEG: how many electrodes are needed?.
Localizing value of electrical source imaging: frontal lobe, malformations of cortical development and negative MRI related epilepsies are the best candidates.
Localization accuracy of distributed inverse solutions for electric and magnetic source imaging of interictal epileptic discharges in patients with focal epilepsy.
30088-8&id=gr1.jpg){kind=link}
30088-8&id=gr2.jpg){kind=link}
30088-8&id=gr3.jpg){kind=link}