Directional recordings of somatosensory evoked potentials from the sensory thalamus in chronic poststroke pain patients

Objective: The aim of this feasibility study was to investigate the properties of median nerve somatosen- sory evoked potential (SEPs) recorded from segmented Deep Brain Stimulation (DBS) leads in the sensory thalamus ( VP ) and how they relate to clinical and anatomical ﬁndings. Methods: We analyzed four patients with central post-stroke pain and DBS electrodes placed in the VP . Median nerve SEPs were recorded with referential and bipolar montages. Electrode positions were corre- lated with thalamus anatomy and tractography-based medial lemniscus. Early postoperative clinical paresthesia mapping was performed by an independent pain nurse. Finally, we performed frequency and time–frequency analyses of the signals. Results: We observed differences of SEP amplitudes recorded along different directions in the VP . SEP amplitudes did not clearly correlate to both atlas-based anatomical position and ﬁber-tracking results of the medial lemniscus. However, the contacts of highest SEP amplitude correlated with the contacts of lowest effect-threshold to induce paraesthesia. Conclusions: SEP recordings from directional DBS leads offer additional information about the neurophys- iological (re)organization of the sensory thalamus. Signiﬁcance: Directional recordings of thalamic SEPs bear the potential to assist clinical decision-making in DBS for pain.


Introduction
Deep brain stimulation (DBS) is an established therapy for several neurological diseases including chronic pain conditions such as central poststroke pain (CPSP) (Krauss et al., 2021). CPSP occurs after vascular lesions that typically affect the central somatosensory system. Among other targets, the sensory thalamus (ventral posterior [VP] according to Hirai and Jones) has been most frequently stimulated to treat chronic neuropathic pain by evoking pleasant paresthesia in order to alleviate pain (Levy et al., 1987;Morel et al., 1997). After more than 50 years of its clinical application DBS for central pain remains a challenging procedure with highly variable outcomes ranging between 25-67% of responders (Katayama et al., 2001;Owen et al., 2006).
One contributing factor might be the difficulty to correctly target the sensory thalamus in these patients due to the limited delineation of the thalamic nuclei on magnetic resonance imaging (MRI) and high inter-individual functional segregation of the sensory thalamus especially in pain patients where functional reorganization can be expected (Lenz et al., 1998).
As a consequence, intraoperative neurophysiological examinations such as evoked potentials have been recognized as an increasingly important tool to characterize the ''physiological" location of DBS electrodes and guide lead placement and postoperative contact selection (Lenz et al., 1998). The latest technological advancement with the introduction of segmented leads offers new opportunities to steer current into certain directions thereby providing further spatial restriction of the stimulation field to maximize the clinical effectiveness while reducing stimulation-induced side effects (Nguyen et al., 2020;Pollo et al., 2014). However, to our knowledge the feasibility to obtain directional somatosensory evoked potentials at the level of the sensory thalamus has not yet been investigated.
Thus, the aim of this study was to investigate the properties of median nerve somatosensory evoked potentials (SEPs) recorded form segmented leads placed within the sensory thalamus. We were interested if there are directional differences of thalamic SEPs and how these are related to the anatomical lead position and tractography findings of the medial lemniscal pathway. Furthermore, were interested in the frequency and time-frequency properties of these SEPs.

Patients and inclusion criteria
For the present study we included four patients (2 females, age ranging from 59 to 70 years) medication-refractory CPSP from of a larger cohort (n = 6) of which clinical data and outcomes were previously published (Nowacki et al., 2022). Directional somatosensory evoked potentials were not recorded in two out of six patients who were excluded from the present study. All patients fulfilled the criteria of definite CPSP according to established grading systems, and were clinically examined for dynamical mechanical (brush) allodynia and abnormal temperature sensibility (Klit et al., 2009). The distribution of subjective pain locations as well as individual findings for brush allodynia, altered temperature perception or heat-and cold-induced allodynia were mapped and documented on standardized pain drawings for each patient and sensory quality as previously published by our group (Nowacki et al., 2022). Patients' clinical characteristics are summarized in Table 1. In all patients, the DBS leads were placed in the sensory thalamus. All patients provided written and informed consent to the procedure and data analysis and the study was approved by the local institutional review board.

Surgical procedure and postoperative testing
Patients were operated under local anesthesia. Leads (8 contacts Boston Vercise Cartesia TM directional leads; Boston Scientific, U.S.A.) were implanted stereotactically using the Leksellframe (Elekta, Sweden) targeting the sensory thalamus (VP) and central lateral thalamus (CL) as previously described (Nowacki et al., 2022) (Fig. 1). Fig. 2 displays the properties of the directional DBS lead. Contact 1 (dome tip contact surface area 6 mm 2 , contact length 1.5 mm) and 8 (contact surface area 6 mm 2 , contact length 1.5 mm) are ring contacts with contact 1 being the most distal. The middle two levels are segmented with three contacts 2/3/4 and 5/6/7 (segmented contact surface area 1.5 mm 2 ). There is a radiopaque marker that aligns with contacts 2 and 5. The outer diameter of the lead is 1.3 mm. Intraoperatively and on the first day after surgery, the electrodes were connected to an external pulse generator and patients were tested for occurrence of paresthesia and side-effects. To this end, each ring-level was tested in a bipolar mode with a fixed frequency of 50 Hz, 200 ls pulse width and increasing stimulation amplitudes starting from 0.5 mA to assess the stimulation threshold for paresthesia. We further tested individual directional contacts on the lead-level that yielded the lowest effect-threshold, again starting from 0.5 mA with 0.5 mA increments until paresthesia could be evoked. Paresthesia thresholds of each ring-mode and individual directional contact were obtained. After electrode implantation, leads were connected to extension cables and to an external pulse generator for 2-4 weeks. After this externalization phase patients decided to have their DBS electrodes connected to an internal pulse generator or -in case they did not profit from the stimulation -to have their DBS leads removed during a second operation under general anesthesia.

Electrode reconstruction and DTI-Tractography
We performed preoperative imaging with a 3-T MRI system (MAGNETOM Trio Tim, Siemens). A standard gadoliniumenhanced T1-weighted protocol (160 sagittal slices, 1-mm thickness) was followed by T2-weighted sequences (FOV 220 mm, acquisition matrix 128 Â 128, repetition time (TR) 2000 msec, and multiple echo times (TE) values ranging from 12 msec to 96 msec in steps of 12 msec) and diffusion weighted imaging (DWI) with multishot echo-planar imaging (number of gradient directions 64, 1.8-mm slice thickness; 62 slices, TR 3300 msec, TE 64 msec, field of view 237 Â 230 mm, b value 1000 sec/mm2). A postoperative high-resolution computed tomography (CT) scan was performed one day after surgery and fused to the preoperative MRI dataset and to reconstruct the electrode artifact with Brainlab Elements TM (Brainlab AG). Reconstructed electrodes were displayed in relation to fiber tracking results of the medial lemniscus. To this end, regions of interest were manually placed onto the primary sensory cortex and the posterior and lateral midbrain tegmentum posterior to the subthalamic nucleus (STN) and medial to the medial geniculate body as identifiable landmarks and setting tracking parameters to an FA-threshold of 0.2-0.3, 50 mm minimum length and 30°maximum angular deviation.
Lead orientation was inferred from intra-or postoperative fluoroscopy to determine the orientation of radio-opaque marker to deduce the direction of each individual directional contact. Reconstructed lead positions were projected onto the stereotactic atlas of Morel to correlate electrode positions with thalamus anatomy and patient-specific fiber-tracking results of the medial lemniscus. To this end, we used the vertical level (axial plane) between the two adjacent directional levels (level two and three) of the DBS lead. The corresponding atlas-level was used and projected onto each patient's individual MRI including the tractographic medial lemniscus reconstruction (Fig. 3).

Intraoperative recordings of SEPs with directional electrodes
SEP recordings were performed during the second stage of surgery (internalization of the implantable pulse generator [IPG] or explantation of the leads) under general anesthesia. Stimulation for SEP of the median nerve was performed with an ISIS IOM System (Inomed, Emmendingen, Germany) using monopolar needles with a square wave pulse of 200 ls duration, rate of stimulation of 2.7 Hz and intensity of 10 -25 mA. Recording was done with scalp cork screws electrodes located in Fz, Cź ,C3 and C4 , derived from international 10 -20 system of electroencephalography (different montages were used, according to the signal-to-noise ratio) (MacDonald et al., 2019;Macdonald et al., 2007), with 100 -150 averages, sweep of 50 -100 ms, and sensitivity of 0.5 -5 lV.
The sampling frequency was 20,000 Hz and a hardware low-pass filter with a cut-off frequency of 5000 Hz was used. Further, we recorded responses to median nerve stimulation from the directional electrodes in the VP with monopolar and bipolar montages using a custom made adaptor to connect the DBS leads with the ISIS. Bipolar recordings were measured across contact 2 to 5, 3 to 6 and 4 to 7. For monopolar montages, each electrode (1-8) was measured with Cz as reference, except for patient 1, where contact 1 was used as a reference against contacts 2-7.

Frequency and time-frequency analysis
Neurophysiological data was analyzed off-line, using custom made Python 3 scripts. Recorded signals were cut to approximately 50 ms (1066 points) for comparability. We analyzed at least three different recordings of both monopolar and bipolar montages for each patient. The raw data was filtered to the following frequency bands, based on the segmentation applied by Pastor and Vega-Zelaya (Pastor and Vega-Zelaya, 2019): 20 -300 Hz, corresponding to the range of low frequency components (LFC) (Note: we use a different terminology than Pastor et al., which will be discussed later) 500 -1200 Hz, corresponding to the range of high frequency components (HFC)  1200 -5000 Hz, corresponding to the range of the very high frequency components (VHFC) In the following, we will use SEP amplitude and LFC amplitude interchangeably, since this approximately corresponds to the frequency band applied to monitor SEPs intraoperatively (usually 30-300 Hz) (MacDonald et al., 2019). A finite impulse response (FIR) filter with Kaiser window with 60 dB attenuation and a width of 0.007 = 700 Hz/NF where the Nyquist frequency (NF) is 10,000 Hz was applied (using the Python package scipy.signal). For all the different components, the amplitude (peak-to-peak) and the latency were determined. Following empirical optimization, the (onset) latency was determined as the time point when the filtered signal crossed the following quantity: M b AE 1r for LFC M b AE 2:5r for HFC and VHFC minus 0.5 ms in the case of LFCs, and minus 1.0 ms in the case of HFCs and VHFCs. Here M b is the mean of the baseline, defined as points 50 to 100 of the filtered signal, and r is the standard deviation of the entire filtered signal. The latencies were doublechecked visually and corrected if necessary. Furthermore, the main frequency of the HFC and VHFC was determined as the weighted average of the Fourier transform applied to the corresponding filtered signal. Finally, using a 1st order Gaussian derivative mother wavelet, a continuous wavelet transform of the raw signal was carried out (using the Python package pywt). The scalograms were plotted over the entire time frame and from 500 to 4000 Hz to capture the dynamics of the high and very high frequency components.

Statistics
Statistical analysis was done with IBM SPSS Statistics Version 28 and was restricted to signal analysis. Data was tested for normality distribution. A student's t-test was applied to compare means of two independent datasets and an ANOVA was applied if more than two groups were analyzed. These groups were mostly either the different patients, the different montages or the different quantities (latency, amplitude, etc.) separated according to frequency components (low, high, very high) within patients (see below). On SPSS, the ''Split file" function was used before carrying out the statistical test, using multiple group levels according to what was being tested. For non-normally distributed data, we used the Kruskal Wallis test.

Seps from directional DBS electrodes in the sensory thalamus
In all patients, we observed a clear directional effect of SEP amplitudes recorded from the DBS electrodes located in the sensory thalamus (Fig. 3). For monopolar recordings the neighboring directional contacts 3 or 6 showed the highest amplitude for median nerve SEP (Supplementary Ta**bles 1 and 2). Accordingly, the bipolar montage of highest SEP amplitude was measured across contacts 3-6 for all four patients (Table 2).

Comparing SEP amplitudes to anatomical electrode position and tractography findings
In all patients, reconstructed electrodes projected onto the VP of the thalamus (Figs. 1 and 3). There was a certain variability in terms of the anatomical location of the stimulating electrode within the VP between patients, a finding that was expected due to adaptations of the definitive target point when considering the somatotopy in the VP to cover the painful area. More specifically, targets were chosen based on the main pain distribution: patients with primarily facial pain (patient 1) were implanted more medially in the medial part of the VP (VPM), whereas patients with a predominant extremity pain (patients 2 to 4) were implanted in the lateral part of the VP (VPL). Furthermore, electrodes projected onto the bulk of the tractography-based medial lemniscus. As can be inferred from Fig. 3, the size and shape of the bulk varied considerably between patients. Of note, in three out of four cases, the directional orientation of those individual electrodes with the highest SEP amplitudes did not point towards the center of gravity of the tractography-based medial lemniscus.

Comparing SEP amplitudes and paresthesia mapping
We found a correlation between the electrode contact of highest SEP amplitude and the contact of lowest effect-threshold to induce paresthesia during intraoperative clinical testing. As shown in Table 1, the most efficient contacts to induce paresthesia in the VP were also the contacts from which the highest SEP amplitudes (Table 2) were recorded during intraoperative median nerve stimulation. This was the case in all four patients.

Frequency and time-frequency analysis
The latencies, amplitudes and main frequencies of bipolar recordings for all patients are summarized in Table 3. Within patients, the following results were obtained. For patients 1 and 3, the latencies of LFCs, HFCs and VHFCs are significantly different (p < 0.001, ANOVA). The amplitude of HFCs and VHFCs are significantly different from the LFCs amplitude for patients 1 (p = 0.005, ANOVA), 2, and 4 (p < 0.001, ANOVA), but the amplitudes of the HFCs and VHFCs are not significantly different among themselves. Across patients, we observed the following. The latency and amplitude of LFCs, HFCs and VHFCs are all significantly different between patients (p < 0.001, Kruskal Wallis). Furthermore, the main frequency of the HFCs and VHFCs of patients 1 and 2 differ significantly from those of patients 3 and 4 (p < 0.001, ANOVA). Overall, patient 2 (thalamic infarct) had the lowest LFC amplitudes. Surprisingly, the amplitudes of the HFCs and VHFCs of patient 2 were higher than the LFC amplitude and similar to the HFC and VHFC amplitudes of patient 1.
The shape of the bipolar LFCs vary considerably from one patient to another (see Fig. 4). As for the HFCs and VHFCs, they come in bursts that are generally restricted to the duration of the LFC. Sometimes there appear to be two subsequent bursts (see Fig. 4, patient 1 HFC, patients 2 and 4 VHFC) or a single burst that diminishes over time (patient 3 HFC and VHFC). Most of the energy of HFC and VHFC seems to be concentrated in the upward deflection of the LFC (see Fig. 4). There do not appear to be separated areas of higher amplitude at any particular frequency, which might indicate that there are no lasting oscillations at any specific frequency (to be discussed in more detail below).
In patients 2, 3 and 4, the latency of the monopolar LFC, HFC and VHFC were significantly different from each other (p < 0.001, ANOVA). Looking at the different montages, significantly different LFC latencies were observed in patients 3 and 4 (p < 0.001, ANOVA), HFC latencies in patients 2 (p = 0.025) and 4 (p < 0.001) and VHFC latencies in patient 3 (p < 0.001). However, there was no clear increasing or decreasing pattern when correlating to montage distance from the tip of the electrode (see Supplementary Ta**ble 3).

Discussion
To the best of our knowledge, this is the first study to examine properties of median nerve somatosensory evoked potentials recorded from segmented leads in the sensory thalamus. We were  Table 3 Mean latencies, amplitudes and main frequencies (+/-standard deviation) of bipolar low frequency components (LFC), high frequency components (HFC) and very high frequency components (VHFC) of bipolar recordings of all 4 patients. Standard deviation is based on 3 to 4 repetitive trials.
Latency ( were determined using postoperative fluoroscopy to identify the orientation of the radio-opaque marker. Using patient specific fibertracking, the position of the medial lemniscus (violet and yellow) was correlated in relation to the ventral posterior thalamus. Corresponding SEP signals are shown on the right side and are connected to each directional contact. Patient 1 (A, B) and 2 (C, D) show a slightly deeper electrode implantation compared to patient 3 (E, F) and 4 (G, H). Right: Low frequency components (red), and high frequency components for the three bipolar montages (2-5, 3-6, 4-7) of each patient. In all patients, the highest SEP amplitude was recorded from the directional contacts 3-6. We found a qualitative correlation between the directional orientation of individual contacts with the highest SEP amplitude and the bulk of the medial lemniscus only in patient 1 (A, B), whereas in patient 2 (C, D), 3 (E, F) and 4 (G, H), the highest SEP amplitude did not correlate with the anatomical projection of the medial lemniscus.
A. Nowacki, D. Zhang, J. Wermelinger et al. Clinical Neurophysiology 151 (2023) 50-58 able to demonstrate the feasibility to carry out directional recordings of SEPs in the sensory thalamus. Amplitudes of SEPs recorded from adjacent directional electrodes differed and showed one predominant direction. Among others, Hanajima et al. recorded SEP from DBS electrodes in response to median nerve stimulation to indicate which contacts are closest to the hand sensory area and which contacts lie above or below this optimal site (Hanajima et al., 2004b). (Hanajima et al., 2004b). Our findings point to a clear directional effect and may confirm the somatotopy of the sensory thalamus. Of note, directional electrodes with the highest SEP amplitudes yielded the lowest threshold to induce paresthesia upon postoperative clinical testing. Furthermore, the spatial orientation of directional SEP amplitudes did neither match the location of the electrodes with reference to atlas-based anatomical subparcellation of the sensory thalamus nor the spatial relationship to fiber-tracking results of the medial lemniscus. Thus, the reconstructed anatomical position and spatial orientation of a directional DBS electrode does not necessarily correlate with the neurophysiological sweet spot. Hence, median nerve SEP recorded from segmented DBS leads constitute a complimentary clinical tool to gather information about the neurophysiological (re)organization of the sensory thalamus beyond anatomical mapping data in patients with chronic pain.

Properties of SEP and high frequency oscillations
In awake patients, the VP is customarily identified by response to light touch or the presence of paresthesia from electrical stimulation (Obwegeser et al., 2000;Wu et al., 2014). In general, during DBS surgery of the VP under general anesthesia, SEP can be recorded either from the DBS lead or from microelectrode recordings (MERs) (Hanajima et al., 2004a(Hanajima et al., , 2004bKatayama and Tsubokawa, 1987). In our study, we solely applied the former method. In the VP response to median nerve stimulation at least two components have been described: a low frequency component (LFC) and a very high frequency component (VHFC) of approximately 1-1.5 kHz (Hanajima et al., 2004b;Klostermann et al., 2002Klostermann et al., , 2000. VHFCs have been claimed to be generated by thalamo-cortical projection neurons and were proposed as a landmark for the VP nucleus (Klostermann et al., 2000;Shima et al., 1991;Yamashiro et al., 1989). Vega-Zelaya et al. described another response termed high frequency oscillations (HFO), corresponding to what we call high frequency components (Vega-Zelaya et al., 2016). Later, Pastor et al. showed that HFO are more specifically detected from the VP, whereas VHFO were recorded also outside the VP (Pastor and Vega-Zelaya, 2019). We chose a different terminology than Pastor and Vega-Zelaya because we are not convinced that the (very) high frequency components of these signals really are oscillations (Luck, 2014) , since there are no lasting distinct regions of high energy in the time-frequency plots (Fig. 4). To our knowledge, we are not only the first group who describe directional recordings of LFCs and HFCs, but also who recorded LFCs and HFCs from the same electrode. Previous groups did use MER for HFC analysis and DBS electrodes for LFC analysis (Hanajima et al., 2004b;Pastor and Vega-Zelaya, 2019).
In the patient with the pure thalamic stroke, we recorded the lowest absolute SEP amplitude value. However, the HFC and VHFC amplitudes were still relatively high. This could indicate that in SEPs recorded from or near the VP, the LFC has a different source than the HFC and VHFC. This is supported by the fact that for patient 1 and 2, the contact of highest bipolar HFC amplitude did not correlate with the directional LFC amplitudes, whereas they did correlate for patient 3 and 4 (Fig. 2). It has to be highlighted that in patient 1 and 2 the DBS lead was implanted just below the VP whereas in patient 3 and 4 it is located at the exact level of the VP (Fig. 1).
The LFC are already generally accepted to be stationary waves from the VP More specifically, (Hanajima et al., 2004b) proposed thalamic SEPs recorded from the DBS electrodes in the STN result from a positive field generated by excitatory postsynaptic potentials in VP neurons in response to median nerve stimulation. For frequency components between 1000 -1500 Hz, thalamocortical projections have been suggested as a source (Hanajima et al., 2004a;Klostermann et al., 2002). The latency pattern between LFC, HFC and VHFC we observed does not seem to contradict this hypothesis. One would indeed expect the (V)HFC to start later than (or at the same time as) the LFC. Looking at monopolar montage latencies, there doesn't seem to be a clear pattern, neither for HFC, nor for VHFC (Supplementary Ta**ble 3). If the (V)HFC were to be the result of an upward travelling wave (along the axis of the DBS lead), one would expect a latency shift: the signal should first arrive at contact 1, then 2-4, then 5-7, and lastly at contact 8. However, the geography of the pathways play a major role in shaping the measured signals which warrants further investigations into the nature and potential sources of (V)HFCs .
It should be noted that it is surprising to observe VHFC in physiological signals recorded from electrodes with a surface area in the order of a few mm 2 . The area is too big to generally detect action potentials from individual neurons and the upper limit for neuron firing rate is about 1 kHz (Kandel et al., 2021). Thus, it is questionable whether the observed VHFC are the result of the same neuron population firing synchronously. Rather, one possible explanation might be that this is the result of different cell populations fire (slightly) asynchronously (which is why we chose to use (very) high frequency components instead of oscillations). In summary, the fact that LFC and (V)HFCs did not correlate in amplitude and especially directionality may be pointing towards a different source of these potentials.

Limitations
First, the low number of included patients is a limitation. In this regards, DBS to treat chronic pain patients is still rarely performed and the use of intraoperative evoked potentials is not at all routine clinical standard. Further, to improve signal analysis in the future, it would be advisable to perform baseline recordings (i.e. before the stimulus) in order to quantify the noise, as well as the changes in frequency and amplitude. Although recordings before the stimulus onset is not standard in intraoperative neurophysiology, we will implement this in the future. It would also be interesting to analyze single sweeps instead of averaged signals.

Conclusion
Our findings provide additional evidence for the somatotopy of the sensory thalamus. SEP recordings from directional DBS leads offer additional information about the neurophysiological (re)organization of the sensory thalamus compared to sole anatomical reconstruction of the electrodes. These signals may serve as a helpful clinical tool to identify the neurophysiological sweet spot in the sensory thalamus of patients with chronic pain. Therefore, SEPs may guide DBS lead placement and facilitate DBS stimulation parameter programming in the future.