Prior to surgery, nTMS mapping of motor function was performed using Navigated Brain Stimulation (NBS) system (Nexstim, Helsinki, Finland). The 3D T1-weighted brain MRI was co-registered to the patient's head using anatomical landmarks and surface matching. EMG surface electrodes (Natus, Paris, France) were placed on deltoid, first dorsal interosseous (FDI), and tibialis anterior (TA) muscles of the right side with an additional large adhesive ground electrode placed over the forearm. Cortical mapping was performed according to nTMS Workshop Report (Krieg et al., 2017), with additional stimulation points at the edge of the tumor, 20 mm deep from the scalp surface (peeling depth). An initial step of rough mapping in the anatomical cortical motor area was done to determine the hand motor hotspot, i.e., the location eliciting the largest motor evoked potentials (MEPs) for the FDI. Since the resting motor threshold (RMT) was higher than usual for the FDI (74%), we decided to perform a motor mapping at only 110% of RMT, the RMT being determined individually for each muscle. As a result, the maximal stimulator output (MSO) used for motor mapping was 81% for FDI, and 55% for both TA and deltoid. The nTMS points eliciting MEP responses were exported at a combined peeling-depth of 15, 20, and 25 mm in the 3D-T1 weighted space.

In addition to an elevated RMT of the FDI compared to deltoid and TA, the nTMS cortical mapping revealed a smaller FDI cortical area compared to the deltoid and TA suggesting an hypo-excitability and/or a loss of corticospinal neurons. Most of the FDI-MEP responses were elicited at the posterior edge of the tumor. Deltoid cortical map (CM) was represented at the normal-appearing posteromedial edge of the tumor in the posterior part of the precentral cortex. TA-CM was also equally represented in the T1 hypointensity parasagittal cortex (superior to the FDI-CM) and in the normal-appearing posterior part of the superior frontal gyrus.

MR imaging was performed on a 3T Verio (Siemens, Erlangen, Germany) imager with a 32-channel head coil. Structural image was acquired using a 3D magnetization-prepared, rapid acquisition gradient echo (MP-RAGE) sequence with the following parameters: TR: 2,400 ms, TE: 3.65 ms, TI: 1,000 ms, flip angle: 8°, voxel size: 1 × 1 × 1 mm³, 190 slices.

Diffusion tensor imaging was acquired with the following parameters: TR: 8,033 ms, TE: 91 ms, 60 transverse sections of 2.5 mm thickness acquired parallel to the anterior commissure/posterior commissure line, number of averages: 1, imaging matrix: 122 × 122 (nominal resolution, 1.967 mm), and acceleration factor: 2. Diffusion weighting was encoded along 12 independent orientations using a b-value of 1,000 mm²/s. For speech tracts (Figures 2g,h), we used the methods described by Wakana et al. (2007) using both seed and target ROIs. For tracking the arcuate fasciculus on the directionally encoded tensor map, a coronal slice was selected at the middle of the posterior limb of the internal capsule. The seed ROI was placed in the deep white matter around the superior longitudinal fasciculus visible as a green triangular shape in the coronal plane. The target ROI was drawn on the axial slice at the anterior commissure level, around the descending portion of the superior longitudinal fasciculus seen as a blue structure. The uncinate fasciculus was tracked by selecting the most posterior coronal slice in which the temporal lobe was separated from the frontal lobe. The first ROI included the entire temporal lobe and the second ROI included the entire projections toward the frontal lobe. For tracking the inferior fronto-occipital fasciculus, the seed ROI delineated the occipital lobe on the coronal slice located between the posterior edge of the cingulum and the posterior edge of the parieto-occipital sulcus. For the target ROI, the entire hemisphere was delineated on a coronal slice at the anterior edge of the genu of corpus callosum. For tracking the inferior longitudinal fasciculus on a parasagittal slice, the posterior edge of the cingulum was located, and a coronal slice was selected at that edge. The first ROI included the entire hemisphere. The second ROI included the entire temporal lobe at the level of the most posterior coronal slice on which the temporal lobe was not connected to the frontal lobe. To dissect the frontal aslant tract, the seed region of interest was located in the white matter of the inferior frontal gyrus and the second region of interest in the white matter of the superior frontal gyrus. To determine the right corticospinal tract, we performed nTMS-guided tractography reconstructions (Krieg et al., 2012) using the nTMS motor ROI as seeds. The tractography was reconstructed with Brainlab fiber-tracking element software (BrainLAB AG, Munich, Germany) using a track angle threshold of 30° and a fiber assignment by continuous tracking propagation algorithm (Mori and van Zijl, 2002) (Figures 2b,c).

Whole brain rs-fMRI data were acquired using a T2^*-weighted GE-EPI sequence with the following parameters: TR: 2,500 ms, TE: 24 ms, 180 volumes, voxel size: 2.5 × 2.5 × 2.5 mm³, 39 interleaved slices, flip angle: 80°, BW: 1,900 Hz/pixel, acceleration factor: 2, acquisition duration: 8 min. During the resting state acquisition, the patient was instructed to keep closed eyes and not to think of anything in particular. For rs-fMRI pre-processing, Matlab R2020b (MathWorks, Natick, MA, USA) and statistical parametric mapping SPM12 software (Wellcome department of imaging neuroscience, London, UK) were used for image pre-processing.

The processing included slice-timing correction, head motion correction, gray/white matter/cerebrospinal fluid structural segmentation. EPI images were coregistered to anatomical images and a spatial smoothing using a Gaussian filter with a full width at half maximum of 8 mm was then performed. Seed-based correlation analyses were performed using the CONN toolbox, which is an open-source Matlab/SPM-based cross-platform software for the analysis of functional connectivity (Whitfield-Gabrieli and Nieto-Castanon, 2012). This toolbox performs seed-based analysis by computing the temporal correlation (bivariate correlation) between the mean blood-oxygen-level-dependent (BOLD) signals from a given ROI to all other voxels in the brain. To remove possible sources of confounds present in BOLD signal data, all fMRI time-series underwent signal compensation from the ventricles, deep white matter and head motion, temporal filtering (0.009–0.08 Hz) on unsmoothed volumes. Seed to voxel connectivity maps were generated using motor and language ROIs as seeds. The motor and language subject-specific ROIs were chosen from the atlases of the CONN toolbox.

This multimodal preoperative mapping showed that motor function was especially at risk in the cortical, subcortical posterior, and subcortical medial areas, in the immediate vicinity of the tumor (Figures 2a–e). The language regions at risk were located on the anterior, posterior, and medial aspects of the tumor (Figures 2f–h).

All pre-operative mapping data were implemented in the Brainlab navigation system in the OR. Craniotomy was performed after the patient was sedated by total intravenous anesthesia based on propofol and opioid infusion, as recommended for performing intraoperative MEP monitoring (Macdonald et al., 2013). In addition, short-duration neuromuscular blocking drugs were only used during endotracheal intubation.

Image guidance by neuronavigation included a correction for brain shift and tissue deformation at two crucial steps of the surgery. For this purpose, intraoperative imaging was acquired with the Airo® mobile CT scanner (BrainLAB) and the BSC was performed with elastic fusion using the Brainlab virtual MRI software (BrainLAB) with biomechanical simulation based on a finite element model taking into account the elastic properties of various brain regions (Riva et al., 2020). The first correction was applied when the tumor bulged through the craniotomy after dural opening: maximal bulge of the brain out of the craniotomy was measured as 6.57 mm (Figure 3a). The second correction was applied when approaching the presumed eloquent brain regions, reaching the medial, anterior, and posterior portions of the resection cavity, i.e., halfway through tumor resection: brain shifts of 5.77, 4.51, and 5.22 mm were found in the anterior portion of the remaining tumor (Figure 3b), for the nTMS-guided corticospinal tract (Figure 3c), and the motor rs-fMRI ROI (Figure 3d), respectively. Relative positions of tumor and motor nTMS ROI were also affected by brain shift (Figure 3e).

After the second BSC, we performed nDES mapping using a disposable bipolar concentric stimulation probe (Inomed, Emmendingen, Germany) and a Nimbus stimulator (Innopsys, Carbonne, France) combined with Brainlab neuronavigation tracking device (Brainlab, Germany) and keeping a bispectral index superior to 55. The DES settings were as usual (frequency: 60 Hz, pulse width: 1,000 μs, intensity: 3 mA) (Pallud et al., 2017). Motor responses were recorded using needle electrodes placed in the left deltoid, biceps, FDI, quadriceps, and TA muscles.

We continued the resection anteriorly guided by brain shift corrected anatomical MRI, speech anterior rs-fMRI data and speech tractography. Regarding the posterior and medial portions of the tumor, we were guided by neuroimaging data as well, but we ultimately respected the functional limits given by nDES and stopped tumor resection even though the parenchyma seemed visually infiltrated.

The updated neuronavigation data with BSC allowed us to directly compare discrepancy between motor cortical mapping obtained with preoperative nTMS and intraoperative nDES (Figure 2e). In this case, after BSC, a distance between DES at 3 mA and the closest nTMS motor positive point was 2.99 mm.

This procedure allowed a total resection of the tumor region with contrast enhancement to be performed, as well as a complete regression of IICP and dysarthria (Figure 4a). However, flair hypersignal with no contrast enhancement was persisting, especially at the posterior aspect of the resection cavity. It could not be removed during this surgery because of functional limits (Figure 4b). Histopathological examination of the tumor confirmed a grade III oligodendroglioma.

Hand paresis worsened slightly during the first post-operative week (2/5) and within 3 weeks it improved compared to the preoperative status (4/5). There was no additional deficit in the immediate postoperative time.

A post-operative nTMS mapping performed 3 weeks after the surgery confirmed the correct functional outcome, with a reorganization of the CM of the three muscles evaluated preoperatively (Figures 4b–d). Compared to preoperative assessment, TA-CM and deltoid-CM were moderately enlarged with a caudal-to-rostral spatial shift. On the other hand, the FDI-CM was reduced with a posterior-to-anterior shift, which justified performing an additional motor mapping for the abductor digiti minimi (ADM), abductor pollicis brevis (ABP) and the extensor carpi radialis (ECR) muscles. The CMs were performed at the same MSO for ADM and ABP than FDI (90%) on one hand, and for ECR than deltoid and TA (50%) on the other hand. The CM of these 3 additional muscles were significantly larger than the FDI-CM, and almost all the responses obtained were obtained for M1 regions distant from the residual tumorous lesion. These results allowed us to inform the patient about the need to focus on the FDI muscle for reinforcement exercises in physical therapy. These encouraging postoperative data led us to discuss the possibility of another resection surgery in the future, in order to prevent, or at least delay, further functional impairment of the patient's right upper limb.

Pre-operative cortical mapping with nTMS provides crucial functional information on cortical regions, specifying the anatomical relations between tumor location and eloquent brain regions. Although nTMS is not suitable to directly investigate deep subcortical structures due to its inherent physical limitations (Lefaucheur and Picht, 2016), subcortical mapping can be enabled by combining nTMS with function-guided tractography reconstructions of the corticospinal tract. Moreover, it is the most precise technique in the postoperative period to plan future resection surgery, giving valuable information about how cortical plasticity and rehabilitation might help to redesign the cortical motor map away from the resection cavity and prevent future relapse.

We did not achieve nTMS-guided speech mapping, although its negative (but not positive) predictive value is well-established in the literature (Picht et al., 2013). In fact, the patient was fatigued by the nTMS motor mapping session, which was performed first. Subsequently, the language assessment was planned but had to be interrupted because the picture naming task was too difficult to perform due to the reduction in the patient's attentional abilities and his fatigue. Nevertheless, it would be useful in the future to perform nTMS language mapping with appropriate patients and implement it in this pipeline.

On the other hand, we assessed the additional information provided by rs-fMRI, although we did not take this information into consideration for performing surgical resection, since its validity to guide surgical resection remains under evaluation (Brennan et al., 2016; Castellano et al., 2017; Ghinda et al., 2018). Indeed, caution should be exercised when rs-fMRI is used for surgical mapping, as the template atlases used for ROI-seed to compute rs-fMRI may have defects due to tumor-induced brain shift and brain plasticity with functional reorganization. We believe that rs-fMRI is still in the research stage and should not be used carelessly for surgical resection.

In fact, there are other non-invasive methods for mapping brain functions that could be implemented in our pipeline. For example, the use of electroencephalography (EEG) or magnetoencephalography (MEG) with source localization of sensory-evoked potentials or magnetic fields to map the sensorimotor cortex appears promising (Vitikainen et al., 2009). These non-invasive techniques could improve the relevance of preoperative mapping and be adjusted in the OR in combination with iCT-BSC.

Spatial precision or resolution of nTMS mapping, tractography reconstructions after DWI, rs-fMRI are respectively on the order of 2–5, 1–3, and 2–4 mm (Glover, 2011; Schmidt et al., 2015; Holdsworth et al., 2019; Marquis et al., 2019) and their agreement with DES was found in the literature to be within 7.7, 8.7–10 mm (Berman et al., 2007; Takahashi et al., 2013; Cochereau et al., 2016). Furthermore, neuronavigation tracking errors in neurosurgery are ~1 mm (Gerard and Collins, 2015) and intraoperative brain shift can be as high as 20 mm and make neuronavigation even less precise and reliable (Gerard et al., 2017). In this case we found brain shifts of about 5 mm for the various anatomo-functional elements at risk (Figures 3c–e). On this basis, it was crucial to combine these preoperative anatomo-functional mapping techniques with intraoperative imaging using BSC since some key structures such as the corticospinal tract can be as thin as 2 mm in some of its portions, as observed in the present case. It was found that, without taking BSC into account, distances to speech tracts shorter than 8–11 mm can be associated with post-operative deficits (Sollmann et al., 2019). Thus, the spatial discrepancy between intraoperative DES and preoperative mapping, depending on the importance of the brain shift, is too high to guarantee a safety margin around the functionally important structures during the resection of brain tumors. This reinforces the need for BSC combined with intraoperative brain imaging to further reduce the risk of post-operative deficits. The use of BSC could significantly increase the spatial accuracy between pre-operative mapping and DES data. Future studies remain to be carried out to compare the respective values of the different preoperative mapping techniques (nTMS mapping, tractography reconstructions after DWI, rs-fMRI).

The use of iCT instead of intraoperative MRI as the intraoperative imaging modality was preferred in our pipeline since iCT is a shorter time-consuming technique and we expected to perform intraoperative brain imaging twice due to IICP. In fact, the surgical interruption when performing iCT-scan lasted 5 min, at both times, whereas intraoperative MRI scanning time ranges typically from 20 to 75 min (Chowdhury et al., 2018) and furthermore is not available in many centers due to its high cost and time-consuming use. There are several intraoperative imaging techniques and models for BSC (Valencia et al., 2012; Morin et al., 2017) but, from economical and practical points of view, the use of iCT may be easier to spread in the neurosurgical community since it can also be used for spine or stereotactic and deep brain stimulation surgical procedures (Scarone et al., 2018; Faust et al., 2019). However, iCT requires that the neurosurgeon be certified for this practice and for radiological protection. In this case, only a radiographer and not a radiologist, has to be available during the surgical procedure. The radiologist only needs to validate the quality and interpretation of iCT at the end of the surgery.

For centers that do not have Brainlab facilities, other methods could be used to compute BSC and implement preoperative mapping. For example, this can be done by re-performing surface mapping for neuronavigation along the exposed cortical veins (which rarely suffer from tumor shift) (Vitikainen et al., 2009) or by using ultrasound or laser range scanner for cortical surface registration to distort the preoperative MRI data according to the brain shift (Mäkelä et al., 2001). The respective performances of these various approaches remain to be compared.

In conclusion, the combination of preoperative mapping advanced techniques, asleep intraoperative motor nDES, and iCT with BSC and elastic fusion to preoperative brain MRI in a patient not eligible for awake craniotomy allowed for complete resection of tumor portions with contrast enhancement without permanent neurological deficits. Here, preoperative nTMS was only used to perform motor mapping, but our proposed pipeline could benefit in the future from preoperative nTMS mapping for speech, visual and cognitive functions.