Open Access
Add to BibSonomy
Abstract
Interictal epileptiform discharges (IEDs) are brief neuronal discharges occurring between seizures in patients with epilepsy. The characterization of the hemodynamic response function (HRF) specific to IEDs could increase the accuracy of other functional imaging techniques to localize epileptiform activity, including functional near-infrared spectroscopy and functional magnetic resonance imaging. This study evaluated the possibility of using an intraoperative multispectral imaging system combined with electrocorticography (ECoG) to measure the average HRF associated with IEDs in eight patients. Inter-patient variability of the HRF is illustrated in terms of oxygenated hemoglobin peak latency, oxygenated hemoglobin increase/decrease following IEDs, and signal-to-noise ratio. A sub-region was identified using an unsupervised clustering algorithm in three patients that corresponded to the most active area identified by ECoG.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Epilepsy is a chronic neurological condition characterized by recurrent seizures caused by excessive brain excitability. For the 30% of patients with epilepsy that do not respond to anti-seizure medications, resective surgery of the epileptic focus can be considered as an alternative treatment [1]. Structural and functional imaging techniques are normally applied before surgery for the purpose of epileptic focus localization, eloquent brain functions mapping, and surgery planning. These include electroencephalography (EEG), magnetic resonance imaging (MRI), functional MRI (fMRI), positron emission tomography, single-photon emission computed tomography, and magnetoencephalography [1]. In some cases, intraoperative electrocorticography (ECoG) is used to guide the extension of the resection. This is conducted with a subdural electrode grid placed directly on the surface of the brain following the craniotomy to allow a 15-30-minute recording in the operating room. Intraoperative ECoG detects epileptiform activity occurring in between seizures i.e., interictal epileptiform discharges (IEDs). While seizures are rare and unpredictable, previous studies reported that brain regions generating IEDs can partially overlap with those generating seizures, therefore the study of IEDs may also provide important information regarding the location of the epileptic focus [2].
Neuronal activity induces local changes in cerebral blood flow through the mechanisms of neurovascular coupling. Blood vessels dilatation near active neurons leads to an increase in cerebral blood volume, a net increase in oxygenated hemoglobin (HbO) concentration, and a net decrease in deoxygenated hemoglobin (HbR) concentration [3,4]. This hemodynamic response function is the basis of many functional imaging techniques such as fMRI and functional near-infrared spectroscopy (fNIRS), which monitor hemodynamic activity as an indirect measure for neuronal activity. In the localization of IEDs, fMRI and fNIRS are usually combined with scalp EEG to determine the timing of IEDs. An event-related design is employed as well as a parametric regression model assuming a fixed shape of the response to IEDs (called the canonical hemodynamic response function, or canonical HRF). However, multiple studies have shown the variability of the hemodynamic response across patients [5,6], age [7], brain areas [8–10], tasks [11] and also in epilepsy [12,13]. For example, characterization of the HRF specific to epileptic spikes using EEG-fNIRS has shown significant inter-patient variability, responses starting before spikes and lasting up to 30 seconds after spikes [13]. The current limitations of EEG-fMRI or EEG-fNIRS restrict the ability to provide accurate estimates of the HRF to spikes. Specifically, fMRI has limited temporal resolution (∼0.5 Hz) and is prone to motion artifacts. On the other hand, fNIRS has a relatively low spatial resolution (∼1 cm) and a low signal-to-noise ratio (SNR) since light has to pass through several layers of tissue before attaining the brain [14]. Moreover, the use of scalp EEG may lower the sensitivity to detect spikes as the electrical signals may be attenuated or cancelled by soft tissue/bone and are frequently degraded by muscle artifacts.
In this paper, we present a proof-of-principle study that demonstrates the application of an intraoperative multispectral imaging system combined with ECoG to measure the hemodynamic response associated with IEDs directly from the exposed cortical surface of the patients. The optical system provides the advantage of having simultaneously a high spatial resolution (∼0.1 mm) and a high temporal resolution (∼10 Hz), as well as no signal contamination from tissue structures. Moreover, the high sensitivity of ECoG allows precise detection of IEDs. Concurrent recordings were conducted in eight patients with epilepsy undergoing resective surgery. The HRF associated with IEDs were obtained for all patients. In three patients, the optical imaging system located a brain area with an increased hemodynamic activity that was consistent with the location of IEDs detected by ECoG. Sampling hemoglobin concentration changes directly from the cortex, this system could be used to characterize patient hemodynamic response to IEDs, which may benefit fMRI or fNIRS analyses by providing more accurate HRF models. We also discuss the accuracy of this approach and its possible clinical uses in epilepsy surgery.
2. Methods
2.1 Patient selection
A total of 12 patients undergoing epilepsy surgery were selected for this study. The medical team performed all standard non-invasive presurgical work-up to identify the epileptic focus. Informed consents were obtained from patients and the Centre Hospitalier de l’Université de Montréal (CHUM) ethics review board approved the research protocol (#14.193). The data of four patients were excluded due to poor optical imaging data quality because of blood accumulation in the surgical cavity (three patients) or no IED at the recording location (one patient), leaving a dataset constituted of eight patients. Patient demographic and clinical information are presented in Table 1. The post-surgery outcome is indicated with the Engel seizure post-operative outcome score, ranging from 1 (seizure-free) to 4 (no worthwhile improvement). The number in parenthesis indicates the number of years since the surgery (corresponding to the optical imaging session) relative to 2022.
Table 1. Demographic and clinical data of all patients included in the study.a
2.2 Intraoperative multispectral imaging
A custom multispectral imaging system connected to a neurosurgical microscope (OPMI Pentero, Zeiss) was used for cortex imaging during surgery [15] (Fig. 1(a)). The imaging system is composed of a snapshot multispectral camera (HSI Snapshot Mosaic sensor, IMEC, Leuven, Belgium) including a 4 × 4 band-pass filters array disposed over a charged-coupled device (CCD) chip. The sensor resolution is 1024 × 2048 pixels, leading to multispectral data cubes of 256 × 512 pixels with 16 spectral channels. The band-pass filters have transmission peaks centered from 480 to 630 nm with an average FWHM of 15 nm.
Fig. 1. Intraoperative multispectral imaging and electrocorticography (ECoG). a) Multispectral imaging system connected to a neurosurgical microscope for video acquisition, b) ECoG electrode grid placed over the cortex during each procedure with active electrodes (in red) and non-active electrodes (in white), with examples of electrical signals for active electrodes and interictal epileptiform discharge identification (red boxes), c) Red-green-blue (RGB) representation of the multispectral data for one patient, with oxygenated (HbO) and deoxygenated (HbR) hemoglobin concentration changes. d) Interictal epileptiform discharges identified with black lines juxtaposed in time with oxygenated and deoxygenated hemoglobin concentration variations for one pixel.
A white light source (Superlux, 300 W xenon lamp) integrated into the microscope was used to continuously illuminate the brain surface during the imaging session. A custom 3D-printed adaptor containing a linear wire-grid polarizer (Edmund Optics, New Jersey, USA) was placed at the distal end of the microscope, with the polarizer aligned with the lamp output. The detection branch of the imaging system included lenses, a short-pass filter (650 nm), a long-pass filter (475 nm), and a second linear polarizer placed in a rotating mount. The position of the second polarizer was chosen to minimize specular reflections (approximately 90° from polarizer #1), by minimizing the camera signal when imaging a reflective material. After adjustments, a sterile drape was installed around the microscope.
The microscope was placed over the patient’s head and the microscope’s zoom and focus were adjusted by the neurosurgeon to maximize the field of view of the area with epileptic activity. The resulting working distance was ∼25 cm and the total microscope magnification was set at 1.5x. The lens system from the adaptor led to a magnification of 4.7, allowing to project an imaging field of view of ∼13 cm2 on the sensor with a corresponding spatial resolution of ∼0.1 mm.
Microscope lamp power was set at 100%. For each patient, the integration time was chosen to maximize signal brightness while minimizing the number of saturated pixels. According to the chosen integration time (ranging from 30 ms to 95 ms), the camera frame rate was set to either 10, 15, or 20 frames per second (fps). Recordings were performed for 8 to 15 minutes.
2.3 Optical data calibration and processing
All data processing was executed after the surgery. Detailed processing methods are described in Pichette et al. [15] but the main steps are summarized here. A spectral calibration procedure was initially performed to account for the system response, based on a reflectance standard (Spectralon diffuse reflectance target, Labsphere) and using an in-house spectral calibration technique validated on optical phantoms. A spatial calibration was also performed with the reflectance standard to correct for vignetting. Surface height variations were within the depth of field of field of the microscope (estimated to ∼ 1 cm) so were not corrected. Pixels with an intensity value equal to the detector’s maximum intensity were associated with specular reflections and were excluded from the analysis. A spatial registration algorithm was applied to the dataset to correct for brain movements due mostly to breathing and heartbeat. The algorithm was implemented using the MATLAB Medical Image Registration Toolbox [16] and was applied to each frame individually. A sub-region within the whole imaged field of view was selected to reduce data size and to minimize computational times. This resulted in images where pixels corresponding to skull were eliminated and where regions containing electrodes were mostly removed.
The modified Beer-Lambert law was applied to the calibrated data to calculate the relative absorbance changes of each data point [15,17]. The absorbance changes were computed relative to a reference image obtained by averaging the entire dataset. The differential pathlength factor used in the modified Beer–Lambert law was computed using estimates from the literature for the absorption and scattering coefficients (µa, µs’) in the brain. Parameter values and equations used to model absorption and scattering coefficients are detailed in [15,18]. A least mean-squares fit was applied to each pixel to find the relative concentrations changes of HbO and HbR using their theoretical absorption spectra [19] (Figs. 1(c) and (d)). Data were filtered temporally to eliminate the DC component and the variations >0.1 Hz associated with physiological components such as heartbeat and respiration rate.
2.4 Electrocorticography and interictal epileptiform discharge identification
Electrocorticography was performed before the optical imaging session. An epileptologist (D.K.N.) identified the brain areas generating the IEDs (Figs. 1(b) and 2(a)). Some electrodes contacts were cut and removed to allow optical imaging without interference from the electrode grid plastic. The electrodes to be removed aimed to maximize the electrode-free field of view for optical imaging while ensuring some of the remaining electrodes would still capture IEDs activity during the imaging session. The decision of which electrodes to cut out was based on four criteria: removing electrodes detecting no IEDs or noise, keeping electrodes with high IED activity, cutting electrodes covering the position of the suspected epileptic focus, and ensuring that the wires connecting the active electrodes to the ECoG system wouldn’t be damaged (Fig. 2(b)-(d)). Optical imaging was then performed with simultaneous ECoG. A time marker was added in the ECoG files to indicate the starting and ending of optical imaging to synchronize the data between the two systems.
Fig. 2. Interictal epileptiform discharge (aka spike) identification and localization procedure using electrocorticography. a) An electrode grid is placed on the cortex to record electrical activity, b) Interictal epileptiform discharges are identified in some electrode positions by an epileptologist, c) the grid is cut to allow optical imaging d) the field of view of the camera is positioned to capture epileptiform activity, e) based on the location of interictal epileptiform discharges, an area is identified as the active area and is registered on the multispectral camera field of view.
After the surgery, an epileptologist (D.H.T.) selected in the recorded intra-operative ECoG data the section associated with the optical imaging session period. The occurrences of IEDs were identified using commercial softwares (Stellate Harmonie or Nihon-Kohden Analyzer). Brain areas with IEDs were identified and registered with the images acquired with the multispectral imaging system to obtain a delineated area with IED activity in the field of view (Fig. 2(e)).
2.5 Average hemodynamic response to interictal epileptiform discharges
For each patient, the average hemodynamic response function (HRF) associated with IEDs was calculated. After applying low-pass filtering at 0.1 Hz and image motion correction (as described in section 2.3), the hemodynamic variations are hypothesized to be a combination of low-frequency activity, resting-state activity, Mayer waves, and epileptiform activity, all having overlapping contributions in the frequency band that was considered. For each pixel, the HbO and HbR sequences were selected from 5s before an IED to 10s after an IED. The 15-second intervals were averaged using a conventional averaging technique. The length of this interval was selected because reported durations of the hemodynamic response are between 5 to 10 seconds following a stimulated event [8]. The 5-second window before IED onset was included because some studies reported hemodynamic changes preceding epileptiform events [13].
2.6 Cortex pixel selection
Because the irritative zone (i.e. area generating IEDs) covered a large surface of the exposed cortex in most patients, the camera’s FOV captured mostly areas generating IEDs and almost no areas without activity. The average HRF across the FOV was then computed by averaging the pixel-wise HRFs across all eligible pixels. Only pixels corresponding to cortical tissue were considered in the averaging. Exclusion criteria included the detection of specular reflections, pixels displaying the electrodes, and pixels associated with large blood vessels. Specifically, 1) specular reflections were identified in the raw multispectral images using an intensity criterion and were masked for all subsequent data processing steps; 2) pixels containing plastic from the electrodes were manually segmented for each patient; and 3) pixels associated with blood vessels were identified based on their reflectance spectrum. For this, a blood vessel mask was defined from a model based on the theoretical reflectance spectra for blood vessels and cortex using optical properties values found in the literature [19,20]. Details of this technique can be found in Laurence et al. [18].
As it appeared that some areas had slightly different HRF profiles, a k-means clustering algorithm (Matlab) was used to detect sub-regions with similar HRF. This step allowed to partition the pixels into cortical regions and to identify pixels with abnormally high intensity, considered to be outliers. Tests with different number of clusters as the input parameter (results not presented here) showed that choosing four clusters allowed us to identify different clusters of outliers scattered across the FOV (2 or 3 clusters depending on which patient was considered) and to identify cortex sub-regions formed by adjacent pixels (1 or 2 clusters). Other input parameters were set to default values. Before clustering, data were normalized using the standard normal variate (SNV) normalization to ensure that the absolute intensity of pixels across the image would not affect clustering, but only the inter-pixels relative amplitude changes.
3. Results
3.1 Average hemodynamic response function to interictal epileptiform discharges
Hemodynamic activity occurring at the surface of the exposed cortex was obtained for eight patients. Given the optical properties of brain tissue in the wavelength detection range used (480-630 nm), the penetration depth of light is limited to a maximum of ∼0.5 mm, thus the reconstruction of the hemoglobin concentration variations is limited to the cortical surface only [21]. The field of view after image cropping, the number of remaining visible electrodes, and the amount of blood accumulated in the surgical cavity varied for each dataset. The clinical data reported in Table 1 illustrates the range of underlying causes and affected brain areas across patients. The occurrence of IEDs also varied across patients with a median time between IEDs ranging from 0.6 to 9.3 seconds (Table 2). Such a wide range of occurrence frequency is representative of the natural inter-patient variability. However, it implies that the number of hemodynamic signal time sequences averaged over to evaluate the HRF is variable across all patients. The number of IEDs used in the calculation of the average HRF varied from 18 to 531.
Table 2. Interictal epileptiform discharge occurrence in patients
An example of the average hemodynamic response to IEDs for one patient (patient #2) is shown in Fig. 3, where maps of HbO and HbR variations are displayed for five representative time points. Variations are observable in most of the FOV but it is hypothesized that the region associated with the highest HbO changes is the region with most of the IED activity. The amplitudes of each HbO and HbR time sequence were normalized between -1 and 1. At t = -5 seconds, the image displays baseline HbO and HbR concentrations associated with low levels of signal intensity in most regions. An increase in the relative HbO concentration is visible in some areas at t = -2 seconds, with a maximum increase localized at the top-center of the field of view. HbO concentration decreased at this location for t = 2 seconds and increased in the bottom-right corner of the image. At t = 6 seconds, only the bottom right corner of the image showed an increase in HbO. The variations in HbR concentration exhibited an opposite behavior, with a decrease in concentration in the middle of the image at t = -2 seconds, followed by a moderate increase. A video of the average HRF shown in Fig. 3 is available on the online version of the manuscript (Visualization 1).
Fig. 3. Average HRF intensity map (HbO at the top and HbR at the bottom) for patient #2 illustrated from -5 to 10 seconds after spike events. Amplitude of the concentration changes were normalized between -1 and 1 for HbO and HbR independently. Pixels corresponding to blood vessels are blacked out. (See Visualization 1)
The optical data quality varied from acquisitions. A quality factor value associated with the raw multispectral imaging data was evaluated for each patient by computing the average value of the first 100 raw frames of each acquisition, divided by their standard deviation at each pixel. The average across all spectral channels was calculated. The quality factor for each patient is shown in Fig. 4(b) in descending order, namely from images with the lower value of photonic noise to those with the highest levels associated with poorer imaging quality.
Fig. 4. Hemodynamic response function to interictal epileptiform discharges and the effect of the quality factor. a) Average hemodynamic response associated with interictal epileptiform discharges for each patient (IEDs occurring at t = 0) displayed in descending order of quality factor. The standard deviation is illustrated by shaded areas., b) Quality factor calculated on raw multispectral images for each patient, c) RGB representation of the multispectral images with high quality factor (patient #1) intermediate (patient #5), and low (patient #8).
For each patient, the pixel-wise HRFs were averaged across all retained pixels after applying the exclusion criteria from section 2.6. This led to patient HbO and HbR hemodynamic responses to IEDs shown (in descending order of quality factor) in Fig. 4(a). For patients #1, 2, 4, and 6, the response showed an increase of HbO and a small decrease of HbR, whereas the opposite behavior is observed for patients #3 and 8. For patients #5 and 7, the standard deviation is large and the behavior of the HRF is less clear. The extremum position varies between patients, appearing before IED onset in two cases (0.7 seconds before for patient #2 and 2.4 seconds before for patient #8).
Figure 4(c) displays an example of the acquired multispectral images for patients #1, 5, and 8, with the highest (P1) and lowest (P8) quality factor. The images are an RGB representation of the multispectral images to illustrate visually the variability in image quality. Exposure times for images of patients 1,5 and 8 were 30, 90 and 30 ms respectively.
3.2 Localization of hemodynamic activity
A k-means algorithm identified outliers (pixels) to be removed from the analysis. For patients #1,2, and 6, two cortex clusters were also identified, corresponding to two cortical sub-regions. The average HRF of the two areas were computed and one sub-region displayed a clear increase in HbO concentration following the IED. This region was selected and compared with the epileptic focus area as identified by the epileptologist. Figure 5(a) displays the maps of the selected sub-region for patients #1, 2, and 6. Figure 5(c) displays the hemodynamic response for pixels included in those clusters for each patient respectively. The HRF show a clear increase in HbO concentration and a decrease in HbR. However, the position of the maximum increase in HbO is different across patients. For patient #1, the maximum increase appears 3 seconds after the IEDs, whereas in patients #2 and 6, the maximum increase appears 1 and 3 seconds before the time of IEDs respectively. A significant HbO undershoot was also seen in patients #1 and 6, reaching its maximum at 10 seconds for patient #1, and at 7.5 seconds for patient #6.
Fig. 5. a) Subregion identified with the clustering algorithm based on the hemodynamic response for patients # 1, 2, and 6 (orange), b) Electrode region detecting IEDs before the imaging session (light green), c) HRF corresponding to pixels identified in a). The two white cables visible in a)-P1 and b)-P1 are depth electrodes.
The approximate area with high epileptic activity as identified by the epileptologist (D.H.T.) based on the ECoG grid is shown in Fig. 5(b) for each of the three patients. We observed a significative overlap between the ECoG localized area and the HRF-based cluster in patient #2, with the major region of overlap in the center of the FOV. For patient #6, the identified cluster partially overlapped with the brain area where the highest level of IED activity was detected (i.e., around ECoG electrodes #20 and 21 located in the center of the FOV). However, some hemodynamic activities, albeit with smaller spatial extent and lower amplitudes, were also seen near electrodes #18 and 19 which exhibited almost no IED activity in the corresponding ECoG time courses. For patient #1, the localization of active electrodes was more difficult as most of the spike activity was localized to be in the deeper structures under the cortical surface (mostly in the right part of the FOV near the depth electrodes). The HRF cluster partially overlapped with the cortical area, which might receive projections of the spike activity from the underlying deep structures. However, a more significant response was observed in the center part of the FOV.
4. Discussion
4.1 Hemodynamic response function to interictal epileptiform discharges
This paper demonstrates that multispectral imaging combined with ECoG can be used to characterize the hemodynamic activity associated with IEDs at the cortical surface. The averaging method used to calculate the HRF considered that the hemodynamic variations were a combination of different types of activity (resting-state activity, low-frequency waves, Mayer waves and epileptiform activity), and that the IEDs contribution would be captured by averaging the 15-second intervals around every IED. Recorded IEDs exhibited a widespread pattern of abnormal electrical activity across the FOV. It was then assumed that the zone displaying IEDs was extensive and covered most of the camera field of view, allowing to characterize the HRF over the whole field of view of the cortex visible in the multispectral images, after removal of noise, electrodes, and saturated pixels. Large areas of hemodynamic activity or activities occurring remotely from the main epileptic site were also observed in other studies and can be explained by widespread IEDs areas or propagation of neuronal activity to secondary sites [22,23]. IEDs median interval was shorter than the 15-second interval selected in the hemodynamic response, meaning that multiple events were averaged in the time frame. This increases the standard deviation of the calculated HRF, but it is assumed that the large number of total events captured still allows distinguishing the overall response to IEDs. The two imaging modalities (optical imaging and ECoG) perform surface measurements only and it is assumed that the errors induced by differences in sampling volumes from the two systems is not significant.
The average HRF obtained in Fig. 4(a) displayed strong heterogeneity between patients, in terms of HRF shapes, oxygenation vs. deoxygenation, and peak latency, which were also reported by other studies [13,24–26]. Common features were observed for patients #1, 2, 4, and 6, where the HRF displayed an increase in HbO concentration and a decrease in HbR concentration, which is consistent with the expected HRF. However, patients #3 and 8 exhibited an inverted response mostly characterized by a decrease in HbO. This has been reported previously, where inverted BOLD response was mostly associated with sustained activity [25]. Patient #3 had the highest number of IEDs (N = 531), occurring every ∼0.6 seconds, thus one interpretation might be that the very high rate of IEDs causes a decrease in oxygenation due to high metabolic demands which exceed the transient increase in local blood flow. The HRF of patients #5, 7, and 8 were associated with a high standard deviation making it challenging to distinguish HbO and HbR responses or to draw any conclusion on the HRF. This might be explained by the low quality factor in the multispectral images for those patients, which made it difficult to separate the two chromophores’ signals. Moreover, patient #7 had the lowest number of IEDs to average (N = 18), which might partially explain the large standard deviation in the HRF.
In terms of latency, the maximum HbO concentration variation observed in our dataset happened at 1.5 ± 4.5 s, with an early response in patients #2 and #8. Early responses were also observed in other studies, where patients showed an HRF preceding IEDs measured by EEG-fNIRS [13] or EEG-fMRI [27] but the physiological origin of this early response is not well understood yet. It was also observed in an EEG-fNIRS study on IEDs that for some patients the HRF lasted up to 30 seconds, with large variations in duration time across patients [13]. A longer selected interval may allow to observe significant changes for patients #5, 7, or 8.
In summary, the large HRF heterogeneity across patients that was observed in this study agrees with the literature and suggests that more extensive studies should be performed to characterize the HRF to IEDs. For example, it would be useful to compare periods with high IED activity and periods without. The dataset acquired in this study displayed large numbers of events and very few periods without IED activity (as seen in Table 2). There were therefore not enough time periods without IEDs to capture baseline activity and compare it to active periods in the same patient with a sufficient statistical power. Future studies should focus on a method to capture both active and non-active periods. No analysis was performed in this study to compare the amplitude of the HRF to the amplitude of IEDs since the amplitude of electrical signals depend on numerous factors and is not reliable. Frequency of occurrence of IEDs should however be studied in terms of HRF shape and amplitude.
One type of functional analysis performed with fMRI or fNIRS involves the deconvolution of an HRF in recorded signals to identify the time and location of neuronal activity. The specificity and accuracy of neuronal activity detection depends on the accuracy of the HRF modeling. However, for locating IEDs in epilepsy, a general HRF is used in most studies even though there are evidence that it is different from normal activity [26]. Choosing patient specific HRFs could improve IED detection using fMRI and fNIRS.
4.2 Localization of hemodynamic activity
The clustering method used in section 2.6 identified cortical sub-regions in the field of view in three patients out of eight. The overlap between the cluster area and the ECoG localized area suggests that the multispectral imaging system can localize hemodynamic changes due to epileptiform activity occurring at the cortical surface. In the other patients, the clustering method suggested that there was no noticeable difference among the HRFs of different regions. This might be due to limitations in signal intensity, or it could also be explained by the rather widespread electrical discharges seen in many of those patients, which implied a potentially large epileptic irritative zone.
Because of instrumental and anatomical limitations, the position of the epileptic focus was not always well centered in the camera’s field of view. In some patients, the epileptic focus was in deeper structures of the brain. In that case, the ECoG measured electrical propagation of the spikes emerging from the focus and reaching cortical surface. However, the propagation of the vascular activity and the neuronal electrical activity may not follow the same trajectory, which may explain the spatial discordance between detected changes in hemoglobin concentrations and spike activity in some patients (e.g., patient #1). Patient #2 represents a case where the extent of the HbO cluster was consistent with the location of the spike electrical activity on ECoG (Figs. 5(a) and 5(b)). This indicates that the detected hemodynamic activity was most probably specific to the epileptic focus. For patient #6, HbO increases were localized both within and outside the brain area where spike electrical activity was observed.
Overall, the results suggested that the employed technique is sensitive in the detection of local hemodynamic changes associated with IEDs, as observed by another study focusing on low-frequency oscillations [28]. More studies should be performed to validate the specificity of focus localization with this technique.
4.3 System improvements
Optical data acquisition showed variability in terms of image intensity and blood accumulation in the surgical cavity, resulting in a difference in quality factor and pixel saturation across patients. Using a more sensitive camera sensor could help to reduce noise since the camera used in the multispectral imaging system had some limitations in terms of dark noise and dynamic range (bit depth of 8) as well as sensor sensitivity, all affecting our ability to limit shot noise (i.e., photonic noise) during acquisitions. Other studies using optical imaging for detection of epileptic activity use mostly two wavelengths [28–32] and some systems up to four wavelengths [33]. The use of a 16-wavelength camera is not crucial; however, it has been demonstrated that using more than two wavelengths helps to reduce errors in chromophore concentrations [34]. The large number of saturated pixels in some patients might be explained by small differences in the alignment of the crossed linear polarizers between acquisitions. Optimizing the alignment method would limit the specular reflections, allow a higher integration time to be used during acquisition, and therefore increase the quality factor. One source of error in the timing precision in our study may come from the manual synchronization between ECoG and multispectral recordings (± 0.5 second) which should be improved.
Other limitations of the study concern the small sample size, the heterogeneity of patient conditions (demographic, type of epilepsy, type and frequency of spikes, etc.) and the lack of control in the surgical parameters (e.g., anesthetics used in surgery may impact hemodynamic responses) increasing the heterogeneity of the measurements. The frequency of IEDs is an important factor that needs to be considered while estimating the HRF and evaluating its accuracy. Within a particular time frame, low-frequency IEDs generate fewer events to be averaged, therefore reducing the confidence interval in the modeled HRF. On the other hand, high-frequency spikes might result in hemodynamic responses in a highly overlapping manner, potentially reducing the accuracy of estimated HRF with the averaging model. However, we have no control over these parameters since IEDs are spontaneous. To overcome this, superior processing methods should be employed, such as a general linear model (GLM) framework. Specifically, the addition of a nonlinear term in the GLM may further improve the accuracy in estimation of HRFs to spikes with high frequency [12]. This was not used in this paper since the focus was to demonstrate the feasibility of using multi-spectral imaging during neurosurgery to detect the hemodynamic response to IEDs but the usefulness of deconvolution and GLM in the estimation of HRF from multispectral data should be evaluated in future studies.
In conclusion, this proof-of-principle study has presented the imaging of hemodynamic changes occurring at the surface of the cortex during epilepsy surgery and highlights the potential of combining ECoG and multispectral imaging to characterize the hemodynamic response function specific to IEDs.
Funding
Canada Foundation for Innovation; Natural Sciences and Engineering Research Council of Canada; Fonds de recherche du Québec – Nature et technologies.
Acknowledgments
The authors would like to thank the medical staff for their help in the operating room and QCVisual for the help with some figures.
Disclosures
The authors declare no conflict of interest.
Data Availability
Anonymized data may be provided upon request.
References
1. J. S. Duncan, G. P. Winston, M. J. Koepp, and S. Ourselin, “Brain imaging in the assessment for epilepsy surgery,” Lancet Neurol. 15(4), 420–433 (2016). [CrossRef]
2. H. M. Khoo, Y. Hao, N. von Ellenrieder, N. Zazubovits, J. Hall, A. Olivier, F. Dubeau, and J. Gotman, “The hemodynamic response to interictal epileptic discharges localizes the seizure-onset zone,” Epilepsia 58(5), 811–823 (2017). [CrossRef]
3. R. B. Buxton, K. Uludağ, D. J. Dubowitz, and T. T. Liu, “Modeling the hemodynamic response to brain activation,” NeuroImage 23, S220–S233 (2004). [CrossRef]
4. B. R. Buchbinder, “Functional magnetic resonance imaging,” in Handbook of Clinical Neurology, J. C. Masdeu and R. G. González, eds., Neuroimaging Part I (Elsevier, 2016), Vol. 135, Chapter 4, pp. 61–92. [CrossRef]
5. G. K. Aguirre, E. Zarahn, and M. D’Esposito, “The variability of human, BOLD hemodynamic responses,” NeuroImage 8(4), 360–369 (1998). [CrossRef]
6. D. A. Handwerker, J. Gonzalez-Castillo, M. D’Esposito, and P. A. Bandettini, “The continuing challenge of understanding and modeling hemodynamic variation in fMRI,” NeuroImage 62(2), 1017–1023 (2012). [CrossRef]
7. M. L. Schroeter, S. Zysset, F. Kruggel, and D. Y. von Cramon, “Age dependency of the hemodynamic response as measured by functional near-infrared spectroscopy,” NeuroImage 19(3), 555–564 (2003). [CrossRef]
8. A. J. Taylor, J. H. Kim, and D. Ress, “Characterization of the hemodynamic response function across the majority of human cerebral cortex,” NeuroImage 173, 322–331 (2018). [CrossRef]
9. M. Li, A. T. Newton, A. W. Anderson, Z. Ding, and J. C. Gore, “Characterization of the hemodynamic response function in white matter tracts for event-related fMRI,” Nat. Commun. 10(1), 1140 (2019). [CrossRef]
10. S. Heinzel, F. B. Haeussinger, T. Hahn, A.-C. Ehlis, M. M. Plichta, and A. J. Fallgatter, “Variability of (functional) hemodynamics as measured with simultaneous fNIRS and fMRI during intertemporal choice,” NeuroImage 71, 125–134 (2013). [CrossRef]
11. M. A. Yücel, C. M. Aasted, M. P. Petkov, D. Borsook, D. A. Boas, and L. Becerra, “Specificity of hemodynamic brain responses to painful stimuli: a functional near-infrared spectroscopy study,” Sci. Rep. 5(1), 9469 (2015). [CrossRef]
12. K. Peng, D. K. Nguyen, P. Vannasing, J. Tremblay, F. Lesage, and P. Pouliot, “Using patient-specific hemodynamic response function in epileptic spike analysis of human epilepsy: a study based on EEG–fNIRS,” NeuroImage 126, 239–255 (2016). [CrossRef]
13. G. Pellegrino, A. Machado, N. von Ellenrieder, S. Watanabe, J. A. Hall, J.-M. Lina, E. Kobayashi, and C. Grova, “Hemodynamic response to interictal epileptiform discharges addressed by personalized EEG-fNIRS recordings,” Front. Neurosci. 10, 1 (2016). [CrossRef]
14. S. Tak and J. C. Ye, “Statistical analysis of fNIRS data: a comprehensive review,” NeuroImage 85, 72–91 (2014). [CrossRef]
15. J. Pichette, A. Laurence, L. Angulo, F. Lesage, A. Bouthillier, D. K. Nguyen, and F. Leblond, “Intraoperative video-rate hemodynamic response assessment in human cortex using snapshot hyperspectral optical imaging,” Neurophotonics 3(04), 1 (2016). [CrossRef]
16. A. Myronenko and X. Song, “Intensity-based image registration by minimizing residual complexity,” IEEE Trans. Med. Imaging 29(11), 1882–1891 (2010). [CrossRef]
17. M. Kohl, U. Lindauer, G. Royl, M. Kuhl, L. Gold, A. Villringer, and U. Dirnagl, “Physical model for the spectroscopic analysis of cortical intrinsic optical signals,” Phys. Med. Biol. 45(12), 3749–3764 (2000). [CrossRef]
18. A. Laurence, A. Bouthillier, M. Robert, D. K. Nguyen, and F. Leblond, “Multispectral diffuse reflectance can discriminate blood vessels and bleeding during neurosurgery based on low-frequency hemodynamics,” J. Biomed. Opt. 25(11), 116003 (2020). [CrossRef]
19. “Optical Absorption of Hemoglobin,” https://omlc.org/spectra/hemoglobin/.
20. S. L. Jacques, “Optical properties of biological tissues: a review,” Phys. Med. Biol. 58(11), R37–R61 (2013). [CrossRef]
21. A. Douplik, G. Saiko, I. Schelkanova, and V. V. Tuchin, “The response of tissue to laser light,” in Lasers for Medical Applications, H. Jelínková, ed., Woodhead Publishing Series in Electronic and Optical Materials (Woodhead Publishing, 2013), Chapter 3, pp. 47–109. [CrossRef]
22. L. Lemieux, H. Laufs, D. Carmichael, J. S. Paul, M. C. Walker, and J. S. Duncan, “Noncanonical spike-related BOLD responses in focal epilepsy,” Hum. Brain Mapp. 29(3), 329–345 (2007). [CrossRef]
23. H. M. Khoo, N. von Ellenrieder, N. Zazubovits, D. He, F. Dubeau, and J. Gotman, “The spike onset zone: The region where epileptic spikes start and from where they propagate,” Neurology 91(7), e666–e674 (2018). [CrossRef]
24. A. Machado, Z. Cai, T. Vincent, G. Pellegrino, J.-M. Lina, E. Kobayashi, and C. Grova, “Deconvolution of hemodynamic responses along the cortical surface using personalized functional near infrared spectroscopy,” Sci. Rep. 11(1), 5964 (2021). [CrossRef]
25. E. Kobayashi, A. P. Bagshaw, C. Grova, F. Dubeau, and J. Gotman, “Negative BOLD responses to epileptic spikes,” Hum. Brain Mapp. 6, 488 (2006). [CrossRef]
26. M. Suh, H. Ma, M. Zhao, S. Sharif, and T. H. Schwartz, “Neurovascular coupling and oximetry during epileptic events,” Mol. Neurobiol. 33(3), 181–198 (2006). [CrossRef]
27. J. Jacobs, P. LeVan, F. Moeller, R. Boor, U. Stephani, J. Gotman, and M. Siniatchkin, “Hemodynamic changes preceding the interictal EEG spike in patients with focal epilepsy investigated using simultaneous EEG-fMRI,” NeuroImage 45(4), 1220–1231 (2009). [CrossRef]
28. Y. Song, J. J. Riera, S. Bhatia, J. Ragheb, C. Garcia, A. G. Weil, P. Jayakar, and W.-C. Lin, “Intraoperative optical mapping of epileptogenic cortices during non-ictal periods in pediatric patients,” NeuroImage Clin. 11, 423–434 (2016). [CrossRef]
29. Y. Ma, M. A. Shaik, S. H. Kim, M. G. Kozberg, D. N. Thibodeaux, H. T. Zhao, H. Yu, and E. M. C. Hillman, “Wide-field optical mapping of neural activity and brain haemodynamics: considerations and novel approaches,” Philos. Trans. R. Soc., B 371(1705), 20150360 (2016). [CrossRef]
30. M. M. Haglund and D. W. Hochman, “Optical imaging of epileptiform activity in human neocortex,” Epilepsia 45(s4), 43–47 (2004). [CrossRef]
31. A. Rayshubskiy, T. J. Wojtasiewicz, C. B. Mikell, M. B. Bouchard, D. Timerman, B. E. Youngerman, R. A. McGovern, M. L. Otten, P. Canoll, G. M. McKhann, and E. M. C. Hillman, “Direct, intraoperative observation of ∼0.1Hz hemodynamic oscillations in awake human cortex: Implications for fMRI,” NeuroImage 87, 323–331 (2014). [CrossRef]
32. K. S. Patel, M. Zhao, H. Ma, and T. H. Schwartz, “Imaging preictal hemodynamic changes in neocortical epilepsy,” Philos. Trans. R. Soc., B 34(4), E10 (2013). [CrossRef]
33. H. J. Noordmans, C. Ferrier, R. de Roode, F. Leijten, P. van Rijen, P. Gosselaar, J. Klaessens, and R. Verdaasdonk, “Imaging the seizure during surgery with a hyperspectral camera,” Epilepsia 54(11), e150–e154 (2013). [CrossRef]
34. C. Caredda, L. Mahieu-Williame, R. Sablong, M. Sdika, J. Guyotat, and B. Montcel, “Optimal spectral combination of a hyperspectral camera for intraoperative hemodynamic and metabolic brain mapping,” Appl. Sci. 10(15), 5158 (2020). [CrossRef]
