Automated Intraoperative Central Sulcus Localization and Somatotopic Mapping using Median Nerve Stimulation

Abstract

Objective:

Accurate identification of functional cortical regions is essential in neurological resection. The central sulcus (CS) is an important landmark that delineates functional cortical regions. Median nerve stimulation (MNS) is a standard procedure to identify the position of the CS intraoperatively. In this paper, we introduce an automated procedure that uses MNS to rapidly localize the CS and create functional somatotopic maps.

Approach:

We recorded electrocorticographic signals from 13 patients who underwent MNS in the course of an awake craniotomy. We analyzed these signals to develop an automated procedure that determines the location of the CS and that also produces functional somatotopic maps.

Main results:

The comparison between our automated method and visual inspection performed by the neurosurgeon shows that our procedure has a high sensitivity (89%) in identifying the CS. Further, we found substantial concordance between the functional somatotopic maps generated by our method and passive functional mapping (92% sensitivity).

Significance:

Our automated MNS-based method can rapidly localize the CS and create functional somatotopic maps without imposing additional burden on the clinical procedure. With additional development and validation, our method may lead to a diagnostic tool that guides neurosurgeons and reduces postoperative morbidity in patients undergoing resective brain surgery.

1. Introduction

Resection of tumors, epileptic foci, or vascular malformations located in the perirolandic cortex carries a serious risk of postoperative injury to the sensorimotor system. The gyral and sulcal anatomy of the brain may drastically differ among subjects, making it difficult for neurosurgeons to recognize different brain regions and identify the position of crucial landmarks such as the central sulcus (CS). Unfortunately, preoperative imaging techniques such as functional magnetic resonance imaging (fMRI) or magnetoencephalography (MEG) are not sufficiently accurate to substantially reduce these risks.

Phase reversal technique (PRT) is a standard procedure used to identify the position of the CS intraoperatively (Allison et al., 1991; Asman et al., 2021; MacDonald et al., 2019; Nuwer, 1986, 2019; Romstöck et al., 2002; Sheth et al., 2013; Simon, 2013). PRT consists of visually detecting specific features of the somatosensory evoked potentials (SSEPs) evoked by median nerve stimulation (MNS) and recording from electrocorticographic (ECoG) electrode grids placed on the surface of the cortex. Specifically, the CS is delineated by identifying the location of the polarity reversal of the voltage peak evoked around 20 ms after the stimulus in the recorded signal. This allows for the identification of recording sites anterior (pre-CS) or posterior (post-CS) to the CS, and hence, of the CS itself (Allison et al., 1989; Balzamo et al., 2004; Cedzich et al., 1996; Goldring, 1978; Jahangiri et al., 2011; Kombos et al., 2000; Simon et al., 2014; Wood et al., 1988).

Direct electrical stimulation (DES) via subdural grids or bipolar electrodes allows for the development of functional brain maps that identify the location of eloquent regions such as the sensory, motor, and language brain areas, which can facilitate safe and optimal resection (Penfield and Boldrey, 1937; Penfield and Jasper, 1954; Penfield and Rasmussen, 1950). However, DES is a time-consuming procedure that may induce seizures or yield misleading non-physiological functional responses (Borchers et al., 2012; De Witt Hamer et al., 2012; Pouratian et al., 2004; Ritaccio et al., 2018). As an alternative, passive functional mapping using ECoG signals represents an emerging technique that has received significant attention as it assesses the physiological response to spontaneous behaviors without the risk of inducing seizures (Brunner et al., 2009), and has been successfully used for intraoperative mapping of eloquent cortex (Breshears et al., 2010; Roland et al., 2010; Taplin et al., 2016; Wu et al., 2017).

PRT and DES are the gold standard procedures for identifying the CS and functionally mapping the surrounding eloquent cortex. While these two procedures yield valid results, their spatial resolution is limited as both PRT and DES require visual inspection, for which time and effort increase with the number of electrodes. Therefore, increasing spatial resolution would put undue additional burden on the clinical procedure, especially in patients undergoing awake craniotomy. In contrast, ECoG-based passive functional mapping does not require visual inspection, and ever-increasing computational power allows for the recording and processing of ECoG signals from hundreds of electrodes simultaneously. With the recent advent of high-resolution grids and biosignal amplifier systems (Viventi et al., 2011), automated procedures may provide an avenue towards rapidly and accurately identifying CS and functionally mapping of the surrounding eloquent cortex.

In this paper, we introduce a new method that complements routine intraoperative MNS (Fig. 1A). This new method is comprised of an algorithm that automatically localizes the CS (Fig. 1C-left) and a technique that identifies the functional region of hand sensorimotor cortex (Fig. 1C-right). Our algorithm localizes the CS by automatically identifying the cortical locations that exhibit phase-reversal in the signals evoked by MNS. Our technique can also be used to create a functional map of hand sensorimotor cortex from changes in the broadband gamma power in the ECoG activity evoked by MNS. Overall, our method could complement current clinical procedures with an automated intraoperative localization tool for rapidly identifying the CS and for functionally mapping of the surrounding eloquent cortex without imposing significant burden on the surgeon or patient.

Figure 1. Current visual inspection-based method and proposed automated method for central sulcus (CS) localization and somatosensory mapping.

(A) The subjects received median nerve stimulation (MNS) during awake craniotomy. (B) Traditional CS localization based on visual inspection of phase-reversal in somatosensory evoked potentials (SSEPs). In this procedure, an electrode strip is placed across the suspected location of the CS while MNS is performed. A trained physician then visually inspects the phase-reversal in the SSEPs to determine the location of the CS. Due to the limited spatial coverage of the electrode strip, this procedure is repeated several times to localize the CS accurately. (C) Proposed automated CS localization and somatotopic mapping. In this automated procedure, an electrode grid is placed within the vicinity of the CS while MNS is performed. Next, an automated algorithm determines the location of the CS from the phase-reversal in the SSEPs (left panel), while an automated method creates sensorimotor maps from the broadband gamma response to the MNS (right panel).

2. Methods

2.1. Subjects and Data Collection

We collected data from 13 human subjects at Huashan Hospital (Shanghai, China) who underwent an awake craniotomy and MNS for the purpose of localizing the CS prior to tumor resection (Table 1). Important to the validation of our method, the tumors did not involve the perirolandic region, and the anatomy was unaffected. This study was approved by the Ethics Committee of Huashan Hospital, Fudan University, Shanghai, China, and we obtained written informed consent from all subjects. An ECoG grid was placed over the CS of each patient (8×8 electrode configuration, 5 mm inter-electrode spacing, 1.8–2 mm exposed diameter for each electrode. PMT® Corporation, Chanhassen, MN, USA; Beijing Sinovation Medical Technology® Corporation, Beijing, China). Needle electrodes (placed at the contralateral and ipsilateral mastoid, N=8 subjects) or a 1×6 strip (placed distant from the CS, N=5) served as reference and ground. To confirm that the grid was placed over hand sensorimotor cortex, we performed standard DES on pairs of ECoG electrodes, or directly on the cortex using a 5 mm wide bipolar electrode stimulator (Epoch XP, Axon Systems, N=7). During DES, the intensity of the 60 Hz biphasic square-wave pulse trains was increased from 1 mA to 2.5 mA until hand movement was elicited. We used a Natus®Quantum LTM amplifier system (Natus Medical Incorporated, Oakville, ON, Canada) with a sampling rate of 2048 Hz to acquire the ECoG signals.

Table 1.

Subject information

Subject	Sex	Age	Hemisphere	Tumor Location	Diagnosis	P1	P2	P3
Sub1	M	19	Right	Temporal lobe	Pleomorphic Xanthoastrocytoma	✓	✓
Sub2	M	31	Left	Frontal lobe	Anaplastic Astrocytoma	✓	✓
Sub3	M	48	Left	Frontal lobe	Astrocytoma	✓	✓
Sub4	F	45	Left	Frontal lobe	Meningioma	✓	✓
Sub5	F	36	Right	Frontal lobe	Astrocytoma	✓	✓
Sub6	F	59	Right	Frontal lobe	Metastatic Adenocarcinoma	✓	✓
Sub7	F	53	Left	Frontal lobe	Anaplastic Astrocytoma	✓	✓
Sub8	F	61	Left	Frontal lobe	Astrocytoma	✓	✓
Sub9	M	39	Right	Insular lobe	Astrocytoma	✓	✓
Sub10	F	31	Left	Frontal lobe	Astrocytoma	✓	✓	✓
Sub11	M	62	Left	Frontal lobe	Astrocytoma	✓		✓
Sub12	F	28	Left	Frontal lobe	Astrocytoma	✓		✓
Sub13	F	30	Right	Frontal lobe	Astrocytoma	✓		✓

P1: Median nerve stimulation (MNS) while awake

P2: Vibrotactile stimulation (VIBS) while awake

P3: Median nerve stimulation under deep sedation

We applied an asleep-awake-asleep anesthesia protocol for this type of awake craniotomy (Lobo et al., 2016; Sewell and Smith, 2019). ECoG signals were recorded while patients were under deep sedation, moderate sedation, or were awake (see Supplementary, asleep-awake-asleep anesthesia protocol). Specifically, we administered MNS to all patients (N=13) while awake. In two control experiments, we administered vibrotactile stimulation (VIBS) while awake (N=10) and MNS during both deep sedation and moderate sedation (N=4). To administer MNS, we first inserted two subcutaneous electrodes at the wrist to stimulate the median nerve contralateral to the recorded hemisphere. Next, we delivered a train of 200us-long monophasic constant-current pulses at a rate of 2.3 (N=9 subjects) or 4.6 (N=3 subjects) or 1 (N=1 subject) pulses per second using the NIM Eclipse® system (Medtronic, Minneapolis, MN, USA). For each subject, we adjusted the MNS current intensity to be slightly above the subject’s individual motor threshold (5.5–15 mA), and recorded the ECoG responses to a train of 53–200 constant-current pulses for each stage of anesthesia. We administered steady-state VIBS to the tip of the contralateral thumb and/or index finger (175 Hz vibration frequency, pulsed, 1–5s on, 1–5s off, linear resonant actuator, 10 mm diameter, C10–100 actuator, Microdrives Ltd., USA). Over the course of 5–10 minutes, our steady-state VIBS yielded a train of 80–160 VIBS pulses per subject.

The neurosurgeons determined the location of the CS by visually inspecting the exposed cortical anatomy prior to placing the ECoG grid and by visually inspecting the SSEPs after performing the MNS procedure. It is important to note that visual inspection of SSEPs was performed independently by two neurosurgeons and without any knowledge of the MNS results in this study. The visual inspection results were further verified by a third neurosurgeon. The independent visual inspection of SSEPs then served as ground truth reference for verifying our automated procedure to determine the location of the CS from the MNS-evoked SSEPs.

Prior to performing our analyses, we visually inspected the recorded signals to rejected any electrodes that did not produce clear ECoG signals, and high-pass filtered the signals at 0.5 Hz to remove slow drifts. Next, we re-referenced the signals to an average of two electrophysiological silent electrodes at the edge of the grid for SSEPs analysis, and re-referenced the signals using a common average reference spatial filter (Liu et al., 2015) for broadband gamma analysis. Finally, we removed line noise and its harmonics using notch filters at 50, 100, 150, 200, and 250 Hz.

2.2. Automatic Localization of Central Sulcus (CS)

In this procedure, we calculated the SSEPs for each electrode as the average across 53–200 MNS evoked potentials. We used the following definition to identify N20/P20 potentials within the SSEPs: N20 potentials exhibit a reproducible negative peak in the 10–25 ms post stimulus and are followed by a large positive deflection; P20 potentials exhibit a reproducible positive peak that arises around the N20 peak in adjacent electrodes. We subsequently used the N20/P20 peaks (so-called phase-reversal) in adjacent electrodes in our automated quantitative algorithm to identify the location of the CS. We performed three steps (Fig. 2): (1) detect evoked potentials; (2) determine whether they belong to N20 or P20 potentials; and (3) reject any evoked potentials that are likely not physiological responses to MNS. The following paragraphs describe these three steps in detail, and the MATLAB code is available in the Supplementary Materials.

Figure 2. Automated procedure to determine somatosensory evoked potentials (SSEPs) polarity and location relative to the central sulcus (CS).

(A) Time periods that exhibit peaks (red) or troughs (blue) in the averaged SSEPs within the 10–25 ms response window are identified. Colored (red/blue) bars represent time periods whose amplitude is significantly different from the baseline period. (B) The peak or trough determines the polarity of the SSEPs, i.e., whether the associated electrodes exhibit pre-CS P20s or post-CS N20s, respectively. (C) Electrodes with a peak or trough occurring outside of the 2 ms confidence interval preceding or following the median peak time (red/blue dashed line) are rejected. The median peak and trough time is derived from all pre-CS electrodes (red dots) and post-CS (blue dots) electrodes, respectively. Results based on SSEPs from Sub1.

Step 1:

We determined the period during which peaks or troughs occurred in response to MNS. For each electrode, we tested the response values of individual time points during the response period (10 ms to 25 ms after the stimulus onset) that were significantly different from the mean values during the baseline period (5 ms to 10 ms after stimulus onset). Specifically (Suppl. Fig. 1A), for individual response time points, we defined a ‘value vector’ as the combination of the response values and the mean values across all trials, and a corresponding ‘label vector’ of the same length (response value = 1, mean value = −1). We then correlated the ‘value vector’ with the ‘label vector’, which provided one Spearman’s R value. We tested the significance of the R value against a shuffled distribution using a randomization test. In detail, the ‘label vector’ was randomly reordered (without replacement) and a new R value computed, and this process was repeated 1000 times. In all cases, we tested for the normality of the distribution using the Kolmogorov-Smirnov test (ks-test). We then determined the probability (i.e., p value) that a given tested R originated from the respective null distribution. For each electrode, this yielded several time windows (Fig. 2A) during which peaks (red) or troughs (blue) occurred (p<0.05, false discovery rate (FDR) corrected for the length of the response period; ks-test, p>0.05).

Step 2:

We used the peak or trough within the first significant window of the SSEP to determine its polarity, i.e., whether it represented a P20 or N20, respectively (Fig. 2B).

Step 3:

We rejected any electrode that did not exhibit a physiological P20 or N20 (Fig. 2C). For this, we determined the median time of peaks and troughs, respectively, and rejected any electrode with peaks or troughs that occurred more than 2 ms before or after. This yielded two groups of electrodes that exhibited either P20 or N20 responses to the MNS. The boundary between these two groups defined the CS (Fig. 3).

Figure 3. Results of localization of the central sulcus (CS).

(A) Results obtained by our automated procedure and by visual inspection for one representative subject. The blue and red shaded areas represent post-CS and pre-CS locations, respectively, as determined by our algorithm. The bold dashed line represents the location of the CS as determined by neurosurgeons through visual inspection of the underlying N20/P20 responses. Intraoperative photographs on the right depict the electrocorticographic (ECoG) grid location (top). As expected, electrical cortical stimulation resulted in a hand motor response only in locations around the CS (bottom). (B) Representative N20/P20 responses for Sub1 (see Suppl. Fig. 5). (C) Results obtained by our automated procedure and visual inspection for all subjects for which the CS was localized by neurosurgeons. (D) An average sensitivity of 89±7% (mean±s.d., N=10) for our automated procedure when compared to traditional visual-inspection-based method, suggests that our automated algorithm accurately identifies the CS.

2.3. Analysis of Broadband Gamma Activity for Functional Mapping

To create a functional map of hand sensorimotor cortex, we needed to determine those electrodes for which MNS evoked a broadband gamma response during the awake stage. We removed the confounding effect of the evoked potential from our responses by subtracting the averaged evoked potential from the individual trials. To extract the broadband gamma response, we applied a zero-phase IIR band-pass filter between 70 and 170 Hz and computed the absolute value of the Hilbert transform on the band-pass filtered signals. To determine which electrodes exhibited an MNS-elicited broadband gamma response, we performed a signal-to-noise ratio (SNR) analysis (Schalk et al., 2007). Specifically, we calculated the SNR as the ratio between the variance of the response period and the average variance of shorter subdivided periods (Suppl. Fig. 1B). We chose −140 to −40 ms and 60 to 160 ms as the response period for the 1 Hz 2.3 and Hz stimulation (−90 to −40 ms and 60 to 110 ms period for the 4.6 Hz stimulation). Next, we subdivided these periods into 5 ms-long non-overlapping bins to calculate one SNR value for each electrode. SNR values larger than 1 are indicative of electrodes whose broadband gamma response was modulated by the stimulus. In contrast, SNR values very close to 1 are indicative of those whose ECoG broadband activity was not consistently modulated by the MNS. We then determined the statistical significance of each observed SNR value through a randomization test. In detail, the broadband gamma signals in each trial were circularly shifted in time by a randomly selected amount. We repeated the process 1000 times, obtaining a null distribution of SNR. We assessed the significance using the same procedure as above. The p values were FDR corrected for the number of electrodes (p<0.01, Fig. 4D,E).

Figure 4. Broadband gamma response to median nerve stimulation (MNS) and vibrotactile stimulation (VIBS) while subjects are awake.

(A) MNS exhibits a stimulus artifact for all electrodes (0–30 ms) as well as a selective functional response (after 60 ms). (B) VIBS exhibits a selective functional response (after 100 ms). (C) Representative functional response induced by MNS (top panel) and VIBS (bottom panel). (D) Comparison of electrodes exhibiting selective functional responses for MNS (red) and VIBS (blue). The diameter of the red/blue circles represents the significance level (p<0.01). Black dots depict electrodes without a functional response. The dashed line indicates CS as determined by neurosurgeons. (E) Comparison of areas with selective functional responses for MNS (red) and VIBS (blue) for all subjects. (F) MNS exhibits an average sensitivity of 92±6% (mean±s.e.m, N=10) when compared with VIBS, suggesting that functional areas identified by VIBS will also be identified by our MNS-based mapping method. (G) This is further confirmed by the high correlation between MNS-induced broadband response and VIBS-induced broadband response (r=0.53 for all subjects, p<0.001. Pearson’s linear correlation). Dots represent electrodes that exhibited significant MNS-induced broadband response.

To visually inspect the MNS-induced activation, we extracted the time-frequency representation for each electrode (Fig. 4A and Suppl. Fig. 2). For this purpose, we applied a short-time Fourier transform from 70 to 240 Hz using a 50 ms-long Kaiser window. We then baseline-corrected and normalized the individual trials by subtracting the average power during each trial’s baseline period (−75 to −25 ms before the stimulus), and dividing it through the standard deviation across all trial’s baseline periods.

2.4. Control Task

In this section, we are interested in verifying the functional relevance of the broadband gamma response to MNS. We hypothesized that the broadband gamma response to MNS represents those functional areas of sensorimotor cortex that receive input from the median nerve. To test this hypothesis, we performed two control experiments (Table 1): VIBS to areas innervated by the median nerve while awake; MNS during deep sedation.

To visually inspected the VIBS-induced broadband gamma activation, we extracted the time-frequency representation for each electrode (Fig. 4B and Suppl. Fig. 3) using the same short-time Fourier transform methods (Sect. 2.3). Next, we used a similar randomization test (Sect. 2.3, Suppl. Fig. 1B) to determine those electrodes that exhibit a significant broadband gamma response to the VIBS in the 0 to 200 ms post-stimulus period (p<0.01, FDR corrected for the number of electrodes, Fig. 4D,E). We chose electrodes that exhibit a significant MNS-induced broadband gamma response to quantify the correspondence between MNS-induced and VIBS-induced broadband responses. In detail, we calculated the mean broadband gamma power over a 100 ms window (60–160 ms after stimulus onset for MNS, 100–200 ms after stimulus onset for VIBS), z-scored across electrodes corresponding to MNS and VIBS, and calculated Pearson’s correlation between MNS-induced and VIBS-induced mean broadband gamma power (Fig. 4G).

For MNS during deep sedation, we were interested in comparing the broadband gamma response during deep sedation to that while awake. We extracted the time course of the MNS-induced broadband gamma response, baseline-corrected each broadband gamma trial by subtracting the average during baseline (i.e., −75 to −25 ms before stimulus onset), and determined the statistical difference between the broadband responses during deep sedation and while awake. Specifically, we divided the response period of each trial (i.e., the 60 to 260 ms after stimulus onset) into 40 bins of 10 ms length with 5 ms overlap between bins, and used the same randomization test (Suppl. Fig. 1A) to determine the statistical difference between the binned broadband gamma responses during deep sedation and while awake (p<0.01; FDR-corrected for all bins and all electrodes of each subject, Fig. 5, Suppl. Fig. 4).

Figure 5. Broadband gamma response to median nerve stimulation (MNS) while subjects are awake or under deep sedation.

(A) Time course of responses to MNS while subjects are awake (red) or under deep sedation (blue). (B) Close examination of the response to MNS for a representative functional related electrode (Sub10, black marker) exhibits the same stimulus artifact (0–30 ms) while awake (red) and under deep sedation (blue), but a statistically significantly attenuated functional response (60–260 ms) under deep sedation (p < 0.01, green bar, broadband response comparison between deep sedation and awake within 60 to 260 ms after stimulus, see Suppl. Fig. 4 for details). (C) Comparison of areas with selective functional responses to MNS (red circles) and areas with statistically significant attenuation during deep sedation (green diamonds) for all subjects. Black dots depict areas without a functional response to MNS. (D) Sensitivity (87±10%) and specificity (86±5%) of attenuation during deep sedation as an index for functionally related areas (mean±s.e.m., N=4). For details see Suppl. Fig. 6C.

3. Results

In this study, we performed post-hoc analysis of ECoG signals recorded from 718 electrodes in 13 patients who underwent MNS in the course of an awake craniotomy (Table 1, Suppl. Fig. 5). We used this extensive data set to develop and validate a new method that can systematically localize the CS and that may help to obtain a functional map of hand sensorimotor cortex.

3.2. Central Sulcus Localization

We first developed an algorithm that automatically localizes the CS based on the PRT by analyzing the MNS-evoked SSEPs. Each electrode was independently assessed to determine its position relative to the CS. Fig. 3 compares the results obtained by our systematic procedure to those obtained by visual inspection. Fig. 3A shows this comparison in detail for one representative subject. The left panel shows the grid schematics, with blue and red shaded areas representing post-CS and pre-CS locations, respectively, as determined by our algorithm. The bold dashed line represents the location of the CS as determined by neurosurgeons through visual inspection of the underlying N20/P20 responses. The right panel verifies these results by comparing them to an intraoperative photograph that depicts the ECoG grid location on the cortex, and the functional response to electrical cortical stimulation. The substantial concordance between our results and those obtained by neurosurgeons indicates that our algorithm can accurately localize the CS for this representative subject. Fig. 3C expands the comparison shown in Fig. 3A to all subjects for which CS was localized by the neurosurgeons (Suppl. Fig. 5). The quantification of this comparison in Fig. 3D demonstrates the high sensitivity (89%) of our automated algorithm to correctly localizing the CS (Suppl. Fig. 6A). In summary, the results in Fig. 3 indicate substantial congruence between the results of our automated algorithm and those obtained through traditional visual-inspection-based MNS.

3.2. Functional Map of Sensorimotor Cortex

After localizing the CS, we were interested in using the MNS-induced responses to create a functional map of sensorimotor cortex. The time-frequency analysis of the MNS-induced responses revealed a non-selective stimulus artifact for most electrodes (0–30 ms), and a selective functional response within the 70–170 Hz frequency range (60 ms after stimulus onset, Fig. 4A for a representative subject, and Suppl. Fig. 2 for all 13 subjects). We determined the sensitivity of our functional mapping technique in producing somatotopic maps by comparing its results to those obtained through the VIBS control experiment (Suppl. Fig. 6B). As the thumb and index finger constitute a subset of the area innervated by MNS, we expected somatotopic maps resulting from VIBS to constitute a subset of the somatotopic maps generated by MNS. Thus, we hypothesized that our method would exhibit a very high sensitivity in this comparison. Indeed, 8/10 subjects exhibited 100% sensitivity, and the average sensitivity across all subjects was 92% (Fig. 4F). Further assessment across electrodes found that 7/10 subjects had a significant (p<0.05) correlation between the MNS-induced and VIBS-induced broadband responses. In addition, we found this relationship to be significant in a group analysis across all 10 subjects (r=0.53, p<0.001, Fig. 4G). In summary, these results suggest that functional areas identified by VIBS will also be identified by our automated MNS-based mapping technique.

3.3. Effect of Sedation

Finally, we were interested in confirming that the broadband gamma responses are indeed induced by MNS and not due to some other non-physiological artifact. For this purpose, we compared the broadband gamma response during deep sedation to that while awake (Fig. 5). We expected a significant broadband gamma attenuation within the functionally relevant areas in deep sedation. This analysis is motivated by prior studies that showed neuronal activity within primary sensory and motor regions under loss of consciousness (LOC) (Durand et al., 2016; Ishizawa et al., 2016; Krom et al., 2020; Lewis et al., 2018, 2012; Malekmohammadi et al., 2018, 2019; Nourski et al., 2017, 2018). For example, Krom et al. found widespread neuronal responses within auditory association cortices during wakefulness. These neuronal responses were attenuated and restricted to primary auditory regions after LOC. At the same time, responses within higher-order auditory-related cortical areas were variable, with neuronal activity being largely attenuated. Ishizawa et al. made similar observations within somatosensory areas. Based on these preliminary studies, we expected that sedation has an attenuating effect on the MNS-induced broadband gamma response. Indeed, the results in Fig. 5A,B and Suppl. Fig. 4 indicate that broadband gamma responses during deep sedation are significantly attenuated for functionally relevant areas (Fig. 5C). Quantification in Fig. 5D revealed a high sensitivity (87%) and specificity (86%) of attenuation during deep sedation as an index for functionally related areas (see Suppl. Fig. 6C for details).

4. Discussion

Intraoperative methods to more precisely identify the different anatomical and functional brain regions are imperative. MNS is a technique that can be used to localize the CS. Despite its utility, traditional visual inspection of MNS results requires substantial time and expertise, and effectively limits the obtainable spatial resolution. In this paper, we present an automated and quantitative algorithm that can rapidly identify the CS without expert knowledge. In addition, the results of our analysis of the evoked broadband gamma responses suggest that MNS could also be used to rapidly create functional maps of hand sensorimotor cortex.

4.1. Physiology of MNS-induced Broadband Gamma Variations

Our method to create functional maps of hand sensorimotor cortex depends on MNS-induced broadband gamma variations. It has been shown that power spectral changes of ECoG signals recorded in the broadband gamma range correlate with population activity and represent the average firing rate of neurons located directly underneath the recording electrodes (Manning et al., 2009; Miller et al., 2014, 2009a; Whittingstall and Logothetis, 2009). Furthermore, it has been shown that both motor activity and tactile stimulation produce broadband gamma responses in both the motor and sensory cortices (Miller et al., 2007, 2009b; Wahnoun et al., 2015). Fukuda and colleagues showed that somatosensory-induced gamma activities in the 100 to 250 Hz range after MNS may represent initial neural processing for external somatosensory stimuli and that these activities may be utilized to localize the primary sensorimotor hand area (Fukuda et al., 2008).

4.2. Broadband Gamma Response of Vibrotactile Stimulation (VIBS)

We validated our functional mapping results by comparing them to those obtained through VIBS, which has been shown to induce broadband gamma activity and accurately map somatosensory cortex (Wahnoun et al., 2015). Furthermore, there is evidence of overlap between cortical areas activated by the VIBS stimulation of median nerve-innervated hand regions and those activated by MNS (Boakye et al., 2000; Sanchez Panchuelo et al., 2016). Indeed, the broadband gamma responses of our VIBS overlap with those of our MNS in this study. It is worth noting though that while they almost completely overlap, the vibrotactile response cannot fully represent the same cortical areas as those mapped by the MNS as we only applied VIBS to a subset (i.e., one to two finger pads) of the hand area innervated by the median nerve. Our results therefore suggest that functional areas identified by VIBS will also be identified by our mapping technique. Future studies could investigate more extensive VIBS for functional mapping.

4.3. Stimulation Artifact as an Inherent Issue

The MNS in our study elicited a non-specific stimulation artifact that made it difficult for us to examine the period up to 30 ms after stimulus onset (Fig. 4A, Suppl. Fig. 2). The presence of a stimulation artifact is an inherent issue present in all MNS studies. Studies that attempt to remove this artifact always face the potential confound that any residual stimulation artifact, high-frequency oscillations (Burnos et al., 2016; Fedele et al., 2017; Sakura et al., 2009), as well as residual N20 or P20 potentials, may easily be confused with physiological broadband gamma augmentations and lead to a false positive in the identification of broadband-gamma-activated electrodes (Amiri et al., 2016; Bénar et al., 2010). Interestingly, our results revealed a significant broadband gamma response to the MNS after 60 ms that is not affected by these artifacts. We further confirmed that this broadband gamma response is indeed induced by MNS and not due to some other non-physiological artifact (Fig. 5). We consequently employed this time interval in our analysis.

4.4. Stimulation Intensity of MNS

The median nerve is a sensorimotor nerve that innervates the forearm and hand. It is well known that MNS elicits evoked responses in sensory and motor cortices with opposite polarities (Allison et al., 1991). Functional MRI demonstrated activation of both motor and sensory areas following electrical and VIBS of peripheral nerves (Ackerley et al., 2012; Boakye et al., 2000; Francis et al., 2000; Sanchez Panchuelo et al., 2016). A recent study that investigated intracerebral recordings in nearly 100 patients found that MNS elicits a broadband gamma response in both sensory and motor areas (Avanzini et al., 2016). To further investigate the sensitivity of the MNS-elicited broadband gamma response within motor areas, they compared these responses to responses elicited by stimulation at intensities 20% below and 10% above motor threshold. Their results did not exhibit any significant difference in the extent that MNS elicited a broadband gamma response in both sensory and motor areas, and thus ruled out the possibility that the strong sensitivity of broadband gamma response in motor areas only depends on stimulation intensity. Our results also showed that MNS elicits a broadband gamma response in both sensory and motor areas (Fig. 4). Furthermore, we verified that these responses are indeed induced by MNS and not due to other non-physiological effects (Fig. 5).

4.5. Effect of Sampling Rate and Notch Filter

MNS-based localization of the CS is typically performed at sampling rates between 2 and 5 kHz (Allison et al., 1989; Fukuda et al., 2008). Our study used a sampling rate of 2 kHz, which is within the range recommended by the literature. Nevertheless, the question arises whether this was sufficient to record the detailed morphology of the cortical SSEPs (i.e., the peaks in the individual P20/N20 response) that define the CS. We performed a control analysis (see supplementary material, Effect of Sampling Rate) to reject the possibility that our sampling rate was too low to record the detailed morphology of the cortical SSEPs and thus identify the location of the CS (Suppl. Fig. 7E).

Another consideration was that notch filters are prone to introducing ringing, i.e., oscillations caused by sharp components, such as the P20/N20 within the cortical SSEPs. We asked whether the notch filters used in this study induced such oscillations, and whether they affected our automated procedure in accurately determining the location of the CS. To address this question, we performed a control analysis (see Supplementary, Effect of Notch Filter) to reject the possibility that notch filter had any adverse effect on the morphology of the P20/N20 within the cortical SSEPs and our ability to determine the location of the CS (Suppl. Fig. 7F).

4.6. Clinical Significance of our Automated Method

Traditional CS localization is based on visual inspection of phase-reversal in SSEPs. Due to the limited spatial coverage of the electrode strip, this procedure is repeated several times to accurately localize the CS (Fig. 1B). Electrode grids have been proposed as an alternative to avoid the re-positioning of the electrode strip and to cut the time necessary to localize the CS by more than half (Nuwer et al., 1992), and such electrode grids are becoming more commonly used nowadays (Kapeller et al., 2015; Parvizi and Kastner, 2018). Furthermore, the required time and the quality of the visual-inspection-based CS localization varies substantially across medical centers, ranging from 1 minute to 20 minutes (Nuwer et al., 1992; Romstöck et al., 2002; Sheth et al., 2013). In combination with electrode grids, our method effectively minimizes the time necessary for the interpretation and reduces the overall procedure to around 3 minutes. Due to the computational simplicity, this duration does not substantially increase with the resolution of the grid and thus the number of electrodes. In contrast, the time required to perform traditional CS localization based on visual inspection of the phase reversal in SSEPs increases exponentially with the resolution of the grid. Thus, our method has the potential to substantially increase the resolution at which the CS can be localized without increasing the duration of the procedure.

Our method also minimizes the necessary time and potential confounds inherent to the communication between the neurophysiologist and the neurosurgeon in the operating room. This communication often becomes a bottleneck and could potentially lead to critical errors due to miscommunication. Visualizing the response to MNS in a form that is readily interpretable by both the neurophysiologist and the neurosurgeon alleviates this issue. Our method addresses this through a topographical representation of the CS-localization (Fig. 3) and functional activation (Fig. 4), that directly corresponds to the cortical electrode array visible to the neurosurgeon.

Our method may also expand the availability of CS localization and functional mapping to a wide range of neurosurgical centers. PRT is a qualitative technique that requires a highly trained and experienced neurophysiologist to perform the MNS and to interpret the resulting evoked potentials. Such experts may not be available widely, effectively impeding the widespread application of this technique. While the proposed automated method may still require the presence of a physician, this physician does not require the same high level of training as a neurophysiologist. Thus, our method may make the localization of CS and mapping of the surrounding functional areas more widely available.

4.7. Ethical Issue of Employing Automated Technology in Healthcare

Despite all these benefits, one major ethical issue remains, i.e., who holds ultimate responsibility for the outcome of an automated procedure? In the absence of regulatory guidelines, this remains an open question (Jobin et al., 2019). However, from the perspective of the patient, the neurosurgeon always holds the ultimate responsibility for the surgical outcome. This conundrum may dampen the enthusiasm of neurosurgeons in adopting automated procedures that can improve their productivity and could potentially expand their reach in benefiting a wider range of patients. This pertains to the broader ethical issues related to employing automated technology in healthcare (Morley et al., 2020). Recently, a group of engineers, ethicists, and social scientists suggested embedding ethicists into the development team as one way of improving the consideration of ethical issues during automated technology development (McLennan et al., 2020). Thus, with appropriate considerations, automated technology can benefit a wide range of patients.

5. Conclusions

This study presents a new automated algorithm to rapidly localize the CS and a technique to rapidly obtain a functional map of sensorimotor areas. Our findings may provide the basis for the development of diagnostic tools that can assist neurosurgeons in the operating room. By providing rapid information about the spatial localization of the CS and the functional mapping of different brain regions at high resolution, this diagnostic tool may reduce postoperative morbidity in patients undergoing resective brain surgery.

Data Availability

Datasets may be provided to interested researchers upon reasonable request to the corresponding author.