In 2009, Picht et al. reported one of the earliest uses of an electromagnetic navigation system to position the nTMS coil in a larger, prospective patient cohort (28). In this and other studies, the accuracy was validated in comparison with direct cortical stimulation (DCS) (mean deviation from DCS 2-4mm) (29–32). A direct comparison between nTMS- and fMRI-determined motor areas with DCS showed better agreement for the nTMS motor mapping (33). When using nTMS, it is important to consider the calculated stimulation accuracy after co-registration with MRI and to accept only a calculated tolerance range of ≤2 mm. Failure to comply has been shown in a study to result in individual cases with significant deviations compared to DCS (34).

Many institutions have explored the clinical benefits of nTMS in surgical planning due to its spatial accuracy in creating three-dimensional functional maps (8, 35, 36). In a study by Rizzo et al., neurosurgeons reported that nTMS motor mapping provided valuable anatomo-functional information in 71% of cases and influenced operative techniques in 29% of surgeries, leading to changes in surgical strategy (37). Frey et al. also observed changes in surgical approaches for over half of patients mapped with nTMS, resulting in a significant increase in gross total resections from 42% to 59% (38). A comprehensive/summarizing meta-analysis by Raffa et al. showed a reduction in postoperative motor deficits (OR = 0.54), an improvement in the extent of resection (GTR OR: OR = 2.32) and an optimization of craniotomy size (-6cm²) and operation duration (-10min) in motor eloquent brain tumors (39).

The idea behind combining nTMS motor maps and DTI tractography was to provide a function-based, individual tractography of the essential tracts of the CST. Conventional tractography algorithms often failed to visualize the CST due to tumor mass effects and peritumoral edema, which has now become possible through the integration of functional nTMS data (40, 41). In detail, Frey et al. defined the term “FA threshold” which enables an individually tailored representation of the CST (40).

In 2014, Conti et al. introduced a technique for somatotopic DTI tractography based on the somatotopic nTMS motor mapping of different muscles (42). The preoperative tractography analyses of 20 patients were integrated into the intraoperative neuronavigation and the accuracy was confirmed using subcortical stimulation. Thus, nTMS-based tractography provides a reliable anatomic and functional characterization of the motor pathway. Rosenstock et al. proposed a manual for standardized nTMS-based tractography confirming the reliability and user-independence of nTMS-based tractography. Even inexperienced users are able to perform the tractography and determine the tumor-tract distance reliably (ICC > 0.9) (43). The superiority in terms of accuracy of nTMS-based tractography compared with conventional tractography was confirmed with direct stimulation (44). In comparison to fMRI as another method for function-based tractography, nTMS-based tractography showed higher plausibility rate, whereas fMRI-based tractography falsely visualized posterior pathways that are presumably functionally more associated with the sensory system (45).

Takakura et al. found a correlation between hotspot-tumor distances and postoperative upper-extremity motor function recovery (46). Patients with greater distances (>10mm) showed better grip strength recovery at 3 months. Krieg et al. demonstrated the prognostic value of nTMS functional mapping in influencing surgical strategy and reducing iatrogenic damage (47, 48). They compared a cohort of 100 nTMS-guided tumor resections with a historical control group and found lower residual tumor rates, improved motor function, smaller craniotomies, and increased eligibility for chemotherapy and radiation therapy in the nTMS group.

Moser et al. and Hendrix et al. further validated the prognostic significance of nTMS positive motor areas anterior to the motor cortex but achieved conflicting results (12, 49). However, when simulating premotor areas care must be taken to ensure that the motor cortex is not stimulated incidentally, which can lead to false-positive results (50).

Several authors investigated the white matter integrity of the CST by analyzing the tumor-tract distance (TTD) as well as the fractional anisotropy (FA) and the apparent diffusion coefficient (ADC). Interestingly, Rosenstock et al. demonstrated that a lower average FA within the affected CST as well as a higher average ADC were associated with deteriorated postoperative motor function (43). In a prospective cohort of 113 patients, no new postoperative motor deficit occurred when the TTD was greater than 8 mm and the precentral gyrus was not infiltrated (27). The relevance of the TTD was confirmed by Sollmann et al. who found that no patient with a TTD ≥ 12 mm suffered from new surgery-induced permanent paresis (51).

Finally, Rosenstock et al. proposed a bicentric-validated risk stratification model using nTMS motor maps and nTMS-based tractography to predict postoperative motor outcomes (short-term and long-term after 3 months) in glioma patients (27, 52). In a study of 278 patients, they found no new permanent deficits when the tumor was > 8 mm distant from the CST and did not infiltrate the precentral gyrus. By assessing the integrity of the CST (by measuring the fractional anisotropy) and of the motor system (by determining the RMT), they could calculate the patients’ individual risk of a new postoperative motor deficit. Further studies demonstrated the clinical and prognostic value of nTMS-based DTI for motor eloquent tumors (53–56). Ivren et al. demonstrated that nTMS-based prediction outperforms an anatomy-based risk assessment (only using structural MRI data) and better predicts the extent of resection (57).

Despite the many published (prospective) studies, no RCT data is yet available since the analysis of the first RCT on the use of preoperative nTMS mapping is pending (NCT02879682).

One of the first experiences of repetitive navigated transcranial magnetic stimulation (rnTMS) was reported by George et al., Herwig et al., and Lioumis et al. for language mapping by inducing temporary disruptions in specific brain regions to identify language-related areas (19, 58, 59). Several stimulation algorithms have been studied over the years, resulting in a recommendation on rnTMS language mapping stated by an international consortium of experts (26): stimulation frequency 5-7 Hz, applied pulses 5-7, picture presentation time 500-700ms, time between picture presentation and stimulation 0-300ms. However, rnTMS language mapping with stimulation frequencies up to 50 Hz was also evaluated in healthy subjects to improve the reliability (60).

In 2013, Picht et al. conducted the first study comparing the use of rnTMS for functional mapping of language areas in combination with direct electrical stimulation (DES) during awake surgery to analyze its reliability in identifying language regions in patients with left-sided lesions (61). The overall evaluation of the following studies showed a high (and therefore clinically significant) sensitivity (range: 63%-97%) and negative predictive value (range: 74%-99%) (62–64). The accuracy appears to be higher for the frontal language regions (Broca’s area), where a sensitivity and NPV of 100% were achieved in two independent studies (47, 61). Thus, tumor resections in rnTMS language-negative areas can be performed with a very low risk for the occurrence of new postoperative language deficits. Some centers even decide not to perform awake craniotomies in these cases, however, the proximity to language-associated tracts must be taken into account (65).

Disadvantages of rnTMS language mapping are the average and highly variable specificity (13%-98%) and positive predictive value (24%-69%). Schwarzer et al. showed a very high rate of false-positive mapping results in patients with neurocognitive deficits and preoperative language impairment, so that rnTMS language mapping should not be performed in cases with baseline error rates >28% (66).

Similar to the nTMS-based DTI FT of the CST, the combination of rnTMS language mapping with DTI studies has been utilized to visualize the language network by emphasizing the rnTMS-associated functional cortico-subcortical tracts. Several studies have explored the integration of rnTMS language mapping and rnTMS-based DTI FT for surgical planning, showing consistent results (67–70). In conclusion, rnTMS-based DTI FT has proven to be valuable in guiding surgical decisions and reducing residual tumor volume (71). However, it should be noted that tractography solely based on rnTMS seeds is not recommended since this will lead to unplausible tractography maps (72). According to the current state of clinically available tractography algorithms, it is rather recommended to first identify pathways based on anatomical ROIs and then integrate the rnTMS information to investigate functional corticosubcortical pathways (67, 73).

In 1998, Pujol et al. reported the first cases of using three-dimensional fMRI mapping to identify the central sulcus as part of preoperative planning for patients with space-occupying lesions (79). The most commonly used task-based motor mapping fMRI methodology is based on conscious, repetitive motor activations in response to the examiner’s instructions (7, 80–82). Such protocols are based on BOLD signal fluctuations between task-induced and control states, whose correlation with intraoperative mapping data has been shown in several studies (83).

Several studies confirmed the clinical utility of tb-fMRI motor mappings, e.g. in selecting the appropriate surgical approach, minimizing craniotomy size (and thus brain exposure), and reducing postoperative neurological complications caused by tumor-induced alterations in expected anatomy and function (84). The studies on neuroplasticity showed varying- and partially contradictory - results. For instance, Baciu et al. (85) and Alkadhi et al. (86) used a block paradigm to study neural plasticity in patients with primary brain lesions, focusing on hand, leg, and mouth movements. They observed that the most common tumor-induced neural plasticity involves inter-hemispheric reorganization, with only few cases of intra-hemispheric reorganization or activation outside the primary motor area. In contrast, Nelson et al. (87) demonstrated varying levels of neural reorganization in the supplementary motor area (SMA), including ipsilateral SMA activation, bilateral activation, and different grades of reorganization. In the small cohort of 12 patients, they found a significant association between the distance between the tumor and SMA region (5mm) and the occurrence of new postoperative neurological deficits (paralysis and speech impairment/mutism).

Although many studies on the clinical applicability of task-based fMRI motor mappings exist, this technology has two significant drawbacks. Firstly, there is no established risk stratification that can differentiate between eloquent-activated areas (where resection leads to neurological deficits) and co-activated areas (where resection results in no or transient deficits). Secondly, several studies have shown that fMRI signal analysis near brain tumors is distorted due to neurovascular uncoupling (i.e. associated with vasogenic edema) and has led to inaccuracies compared to the gold standard (DCS) (88–93). Thus, the motor cortex could not be identified in 18% of patients and the distance between the DCS hotspot and fMRI hotspot was frequently >1cm, which hindered adequate surgical planning (94–96). Ultimately, the sensitivity and specificity of fMRI were 61.7% and 93.7% (94, 97, 98). In DCS-based comparative studies, fMRI was significantly inferior compared to nTMS in terms of identifying motor eloquent areas (33, 95, 99). In an effort to improve subcortical spatial orientation during surgical resection, fMRI-identified areas were used as ROIs for DTI tractography of the corticospinal tract (100, 101). In a comparative study, fMRI-based tractography was significantly inferior to nTMS-based tractography in terms of plausibility and occurrence of aberrant fiber pathways. Particularly when the tumor was in close proximity to functional areas, changes in BOLD signal physiology occurred, highlighting nTMS-based tractography as the method of choice (45).

Dierker and colleagues were among the first to compare rs-fMRI with tb-fMRI acquisition (102). By detecting task-negative states, rs-fMRI enables researchers to identify spontaneous fluctuations in BOLD signals, forming resting-state networks (RSN) (103). The main benefit of rs-fMRI is the ability to generate functional maps even in uncooperative individuals (such as children or incompliant adults), as long as they remain calm during the scan (104–108). rs-fMRI tends to additionally detect network-based associated regions (such as the sensory cortex) (109), resulting in the detection of larger areas but also raising further questions about their surgical relevance and usability.

The aforementioned disadvantages of tb-fMRI are even more pronounced here, as rs-fMRI mapping relies essentially on identifying preserved anatomical landmarks that may be altered by the lesion. Furthermore, the acquisition times for rs-fMRI can still be too long at times, although interesting new advances through machine learning (ML) and artificial intelligence show promising results (110).

Despite the complexity of the language network, the initial experiences date back to 1999 by Ruge et al., where a good concordance between tb-fMRI activated areas and DCS was found in 5 patients (111). Current standard paradigms include picture-naming tasks and word-listening tasks, which are also utilized in assessing neuroplasticity (112, 113). Although many studies have reported positive experiences with fMRI for investigating the language network (13, 114, 115), very similar limitations exist as with fMRI motor mapping. Particularly when the tumor is located very close to eloquent areas (<1cm), involves high-grade gliomas with blood-brain barrier disruption and perifocal edema, or well-vascularized tumors, authors have found significant inaccuracies (sensitivity: 37.1%, specificity: 83.4%) (116, 117). In another study involving 50 patients with left-hemispheric gliomas, in some cases, a right-hemispheric language dominance was mistakenly indicated, which was explained by a reduced perilesional fMRI signal due to lesion-induced neurovascular uncoupling (118).

Studies on rsfMRI have shown that even uncooperative patients could be examined, leading to the detection of co-activated areas and thus overcoming this intrinsic limitation of tb-fMRI. However, this also complicates the interpretation of results, as it remains difficult to differentiate between eloquent and non-eloquent, co-activated areas (119–121). rsfMRI analyses should be interpreted with caution, especially when tumors alter the anatomical landmarks as well as the functionality of the RSN, leading to incorrect results (122).

The first clinical experience with magnetoencephalography (MEG) dates back to 1995, when Gallen et al. attempted to utilize somatosensory evoked potentials to predict the pattern of phase reversals observable intraoperatively on the cortical surface in six patients with various pathologies. They identified a correlation across the entire cohort, with an average deviation of 8 mm compared to the intraoperative findings (126). In 1996, Rezai et al. reported on the integration of MEG-derived brain mapping with various imaging techniques to develop a preoperative algorithm for navigating during the resection of eloquent lesions in the sensorimotor cortex (127). Schiffbauer et al. used MEG to investigate the spatial relationship between functional eloquent areas and tumors, finding functional activity in 18% of grade 2 tumors, 17% of grade 3 tumors, and 8% of grade 4 tumors (128).

In numerous studies, MEG results have been compared with fMRI analyses and validated in a few studies using intraoperative DES. Overall, there was good agreement (and partial superiority) with fMRI analyses (129–131), but a mean deviation of DES-identified motor hotspots of 12.5mm, which poses a significant limitation for neurosurgical operation planning (132–135). This spatial deviation poses a particular challenge for precise neurosurgical planning. MEG localization often relies on the mapping of somatosensory evoked fields (SEFs) rather than signals from the motor cortex. SEFs localize the somatosensory cortex, with sources located deeper in the central sulcus, while direct electrical stimulation (DES) targets the crown of the precentral gyrus to identify motor hotspots. This anatomical variance between somatosensory and motor cortex activation contributes to the observed spatial discrepancies and must be taken into account (136, 137). The reliability of motor mapping for planning brain tumor resections is not compromised in centers with experience in MEG analysis, as the deviation between the activity shown by MEG and the DES stimulation was generally ≤10 mm (133).

Tarapore et al. compared nTMS with MEG and DES for cortical motor mapping in patients with malignant lesions near the central region. While a direct comparison between MEG and DES was not conducted, the mean distance between nTMS and MEG imaging motor sites was 5 mm, and between nTMS and DES motor sites was 2 mm, suggesting a strong alignment between MEG and nTMS (34). In one case of the study, no MEPs could be elicited by preoperative nTMS or intraoperative DCS; however, there was motor-functional MEG activity within the lesion. Postoperatively, a permanent motor deficit was evident in this case. This individual case suggests that MEG may be valuable in situations where neither nTMS nor DCS is able to induce MEPs.

In a study of 119 patients with gliomas, the relative spatial relationship between the tumor and eloquent brain areas (specifically sensory-motor, visual, and language areas) was analyzed (138). In high-risk cases with distances <5mm, surgery was avoided, which occurred in 46% of patients. Although the rate of postoperative neurological deficits was low at 6%, it raises questions about whether some non-operated patients may have been suitable for resection, potentially missing the opportunity for longer overall survival.

In a feasibility study, the language associated areas detected by MEG were used as ROI and combined with diffusion tensor imaging (DTI) to visualize the arcuate fasciculus in patients with lesions in language regions. However, the clinical relevance and benefits for surgical planning and the correlation with postoperative outcomes have yet to be investigated (139).

In contrast to MEG motor mapping, there are very few studies available on MEG-based language mapping, which primarily focus on interhemispheric language lateralization. In one of the initial studies, Grummich et al. combined fMRI and MEG to investigate language localization in 172 patients with brain lesions (140). They found a high level of agreement between the two techniques in identifying language sites (77% congruence), with only a small percentage showing differences (4%). MEG mapping was shown to be superior in nearly half of the cases (in which the BOLD signal was suppressed, particularly in glioma patients). Another study investigating language localization was conducted by Szymanski et al. in 2001, who demonstrated a general left asymmetry in right-handed neurosurgical patients through MEG mapping, as well as a right asymmetry in two patients with confirmed right hemispheric language dominance via the amobarbital test (141).

While the Wada test is nowadays used by only 12% of epilepsy centers for language lateralization assessment, it has been considered the “gold standard” for preoperative determination of language dominance (142). Doss et al. validated MEG-based language localization with the Wada test, reporting a relatively low concordance rate (69%) between the two methods (143). Ota et al. (144) compared MEG and fMRI with NIRS (near-infrared spectroscopy) and the Wada test for language lateralization, finding that fMRI and NIRS had higher sensitivity and specificity in patients with typical hemispheric dominance, while NIRS showed better specificity in patients with right language lateralization (144). However, the authors emphasized the individual, high technical requirements (device differences, patient characteristics such as brain edema and movement artifacts) associated with MEG, fMRI and NIRS, which make (routine) use difficult (145). In a comparative study between MEG and nTMS language mapping for determining language dominance, a 64% agreement was found. However, no validation with the WADA test was performed which limits the significance of the study (146).