Evolution of the approaches to target identification for therapeutic repetitive transcranial magnetic stimulation

Full Text

Introduction

Transcranial magnetic stimulation (TMS) is a method of non-invasive brain stimulation using an alternating magnetic field, which allows for non-invasive and painless depolarization of neurons at the stimulation site, followed by propagation of excitation along neuronal pathways [1, 2]. When using repetitive TMS (rTMS) protocols, long-term changes in the activity of the stimulated area and associated neuronal networks develop due to the induction of processes similar to long-term potentiation and depression [3–6]. This enables the use of rTMS for therapeutic neuromodulation in various nervous system disorders. The most widely used protocols in clinical and research practice include standard high- and low-frequency rTMS protocols, as well as theta-burst stimulation protocols [1, 2, 7].

According to the recommendations of an international expert group (2020), the most convincing level of evidence for the efficacy of rTMS (Level A — definitely effective) has been obtained for its use in treatment-resistant recurrent depression, neuropathic pain, and post-stroke arm paresis within the first 6 months. For many other diseases and syndromes (motor impairments and depression in Parkinson’s disease (PD), lower limb spasticity in multiple sclerosis, aphasia, Alzheimer’s disease, etc.), the levels of evidence for efficacy are defined as B (probably effective) or C (possibly effective) [7].

In recent years, rTMS has been increasingly used in real-world clinical practice due to the accumulation of evidence and its inclusion in clinical guidelines for several conditions. At the same time, the limitations of standard rTMS protocols have become increasingly evident: in most cases, they produce only small or moderate effects that are highly variable [7–11]. Consequently, approaches to enhance the targeting and personalization of rTMS protocols are actively developing, which ideally should provide provide brain stimulation in the right location and at the right time with adequate values of all parameters [12–17]. The selection of the target and coil positioning method is a crucial component in this context, as it determines the pool of neurons whose activity is modulated [12–15].

The target area for rTMS is determined by the understanding of the pathophysiology of a specific disease and syndrome, considering potential network effects that enable modulation of activity in brain regions distant from the stimulation site but functionally connected to it [7, 16]. In standard rTMS protocols, the target is defined anatomically as a relatively large cortical area (e.g., dorsolateral prefrontal cortex (DLPFC), primary motor cortex (M1), posterior parietal cortex, etc.). Most rTMS protocols use figure-eight coils, which provide relatively focal stimulation of the convexity cortex within approximately 3–5 cm² and with a penetration depth of 2–3 cm from the scalp [18–20]. The standard approach is based on coil positioning using external landmarks or craniometric measurements, accounting for averaged data on the localization of cortical projections on the head surface [21, 22]. The decisive technological breakthrough in this field has been the introduction of neuronavigation systems. The coil can be positioned with high precision relative to the target cortical area only with navigational control [23–26]. This, in turn, has driven the development of numerous methods for determining optimal rTMS targets using the rich arsenal of modern neuroimaging techniques [14, 27–32]. Conceptually, neuronavigation has enabled the shift toward defining personalized rTMS targets — one of the most promising and rapidly evolving research areas of non-invasive brain stimulation research in recent years [12, 13, 19, 27].

The aim of this review is to examine the main approaches to defining rTMS targets and coil positioning, and to describe their methodology, evidence base, key advantages, and limitations.

Standard approaches to target identification for rTMS and coil positioning

Standard approaches involve positioning the coil relative to large anatomical targets (M1, DLPFC, auditory cortex, etc.) without neuronavigation. The specific coil position on the scalp can be determined relative to the M1, by external landmarks, or by the placement of electrodes of the ‘10-20’ system for EEG [12, 21, 22].

M1 can be readily localized using single-pulse TMS and recording motor evoked potentials (MEPs) from the target muscle. The target for rTMS is the “hot spot” of the target muscle — the coil location at which MEPs with maximum amplitude are recorded. In motor disorders associated with corticospinal tract lesion, stimulation is most commonly applied to the area of the motor cortex somatotopically corresponding to the location of the symptoms [7]. In M1 rTMS for patients with disorders not involving the corticospinal tract (e.g., neuropathic pain, PD), in recent years the cortical representation of hand muscles has typically been chosen as the rTMS target due to the ease of locating it and the lack of data supporting higher efficacy of stimulating the somatotopically corresponding symptomatic area [33–36]. The rTMS therapeutic effects in these cases develop due to the spread of excitation along fibers connecting the motor cortex with other brain regions (thalamus, striatum, cingulate gyrus, etc.), rather than along the somatotopically organized corticospinal tract [37, 38].

For several cortical areas, stimulation targets are defined relative to M1. For example, DLPFC under the standard approach is located 5–6 cm anterior to the hand muscle hot spot [22, 39], premotor cortex — 2 cm anterior and 1 cm medial to this point [40, 41], supplementary motor area — 2–3 cm anterior to the leg muscle hot spot [42].

Another common approach involves defining rTMS targets on the scalp relative to the international 10-20 EEG electrode system [20]. Target location may be specified by individual electrode positions or calculated relative to multiple electrodes. For instance, left and right DLPFC correspond to F3 and F4 electrodes respectively [39], left and right posterior parietal cortex — P3 and P4 electrodes respectively, left auditory cortex — midway between T3 and P3 electrodes [20]. An alternative method for auditory cortex localization positions the coil 2.5 cm superior along the T3–Cz line and 1.5 cm posterior perpendicular to this line [43]. The table presents standard approaches for locating common rTMS targets.

 

Standard methods for identifying therapeutic TMS targets (adapted from [36])

Target	Localization methods	Examples of conditions where stimulation is applied
M1	·                  Point with maximum MEP amplitude from the target muscle; ·                  5 cm lateral and 1–2 cm anterior to Cz electrode (cortical representation of hand muscles) or 0–2 cm posterior to Cz (cortical representation of leg muscles); ·                  C3 (left M1) and C4 (right M1) electrodes of the 10–20 system	Post-stroke motor deficit, neuropathic pain, fibromyalgia, PD, post-stroke dysphagia, spasticity in multiple sclerosis, motor deficit and spasticity in spinal cord lesions, chronic disorders of consciousness, migraine
Premotor cortex	·                  2 cm anterior and 1 cm medial to the hot spot for hand muscle in M1	Writer’s cramp, post-stroke motor deficit
Supplementary motor area	·                  2 or 3 cm anterior to the hot spot for leg muscle; ·                  anterior to Cz at 15% of the nasion-inion distance (pre-supplementary motor area)	PD, obsessive-compulsive disorder
DLPFC	·                  5, 5.5, or 6 cm anterior to the hot spot for hand muscle in M1; ·                  F3 (left) or F4 (right) electrodes of the 10–20 system	Depressive episode, schizophrenia (negative symptoms), post-traumatic stress disorder (PTSD), cognitive impairment, migraine, insomnia
Orbitofrontal cortex	·                  Fp1 (left) or Fp2 (right) electrodes of the 10–20 system	Obsessive-compulsive disorder
Posterior parietal cortex	·                  P3 (left) or P4 (right) electrodes of the 10–20 system	Cognitive impairment, post-stroke spatial neglect (neglect syndrome)
Right inferior frontal gyrus	·                  2.5 cm posterior to the lateral canthus along the line connecting the lateral canthus and tragus, then 3 cm perpendicularly upward; ·                  F4 electrode of the 10–20 system	Post-stroke aphasia
Auditory cortex	·                  For tinnitus: 2.5 cm upward along the T3–Cz line, then 1.5 cm posterior perpendicular to this line; ·                  for verbal hallucinosis in schizophrenia: midpoint between T3 and P3 electrodes of the 10–20 system	Schizophrenia (verbal hallucinosis), tinnitus

 

The greatest number of approaches have been described for determining coil localization during rTMS of the DLPFC in patients with depression (Fig. 1) [39]. In the mid-1990s, the 5 cm rule was proposed and remains the most widely used method to date [44–46]. This approach was used in the majority of large clinical trials that formed the evidence base for high-frequency rTMS of the left DLPFC in treatment-resistant depression, including the study by J.P. O’Reardon et al. [47], on the basis of which the method was approved by the Food and Drug Administration (FDA) for this condition. At the same time, it was shown that when using the 5 cm rule, the target is located within the DLPFC in less than half of cases [48, 49]. An association between a more pronounced antidepressant effect and a more anterior and lateral target location was identified, leading to the proposal of locating the target in the DLPFC at 5.5 or 6 cm anterior to the motor cortex [39, 50, 51]. As an alternative, a method for determining the left and right DLPFC based on the location of the F3 and F4 electrodes, respectively, has been described [39]. Subsequently, the localization of the F3 electrode, which requires a series of craniometric measurements within the classical 10–20 system, was simplified and automated using the F3-Beam approach [52].

 

Fig. 1. Examples of target localization for rTMS of the DLPFC.

1 — F3 according to “10-20” system; 2 — 5 cm; 3 — 5.5 cm; 4 — 6 cm; М1 — primary motor cortex. Visualization was performed using the Visor2 neuronavigation software (Ant Neuro).

 

There are still no data on differences in the antidepressant effect of rTMS to the DLPFC in depression depending on target localization methodology within the standard approach, which should be interpreted considering the fact that this issue has essentially not been studied in controlled trials. In this regard, one can note a small study showing no statistically significant differences in the antidepressant effect of high-frequency rTMS to the left DLPFC when targeting using the 5.5 cm rule versus the F3-Beam method [53]. At the same time, target localization differs substantially when using various standard approaches; for example, targets identified by the F3-Beam method are on average 2.6 cm more anterior and more lateral compared to targets identified by the 5.5 cm method [51].

Overall, within the standard approach, targets for rTMS are approximated based on averaged statistical data about the projection of the corresponding cortical area onto the scalp surface, obtained from small groups of healthy individuals. Determining target locations based on the 10–20 electrode system allows for partial accounting of individual differences in head size, since electrode placement itself is determined through craniometric measurements [21]. However, it is evident that the standard approach cannot account for individual anatomical and topographical features of the cerebral cortex, often resulting in stimulation of non-target cortical areas [12, 16, 48, 49]. Moreover, the standard approach cannot distinguish between non-motor targets within a single anatomical region. The exception is the M1, and even for this area, differentiation is only possible within the somatotopy of corticomuscular projections, but not for other connections of the M1 [37, 38].

Neuronavigational approaches

Overview

The introduction of navigated TMS systems in the early 2000s served as a technological prerequisite for virtually unlimited new approaches to more precise targeting for rTMS, enabling coil positioning relative to a given target with extremely high (within a few millimeters) spatial resolution [23, 24]. With navigated TMS, by correlating surface anatomical structures with their location on an individualized 3D head model obtained via structural MRI, it is possible to precisely determine the spatial position of the coil and the induced electric field relative to the cerebral cortex of a specific individual (Fig. 2) [26]. More complex variants involve various mapping methods for more accurate identification of the target area within a single anatomical area; most commonly, resting-state fMRI or task-based fMRI is used for this purpose, less frequently diffusion tensor tractography, positron emission tomography, near-infrared spectroscopy, and other methods [12, 16, 28, 32].

 

Fig. 2. An example of visualization of a coil for stimulation and induced electric field using the Visor2 neuronavigation software (“Ant Neuro”).

 

In addition to precise coil positioning, neuronavigation enables objective monitoring of its position during the session. It has been shown that fluctuations in coil position and orientation are a significant source of variability in MEP amplitudes [54]. This factor may be even more critical for variability in therapeutic rTMS effects, given the longer procedure duration and inevitable head movements even with the coil in a fixed position. Studies demonstrate that in healthy individuals, navigated low-frequency rTMS of the motor cortex produces stronger behavioral effects in psychomotor tests compared to non-navigated rTMS [55]. Since the target in the motor cortex was identified via localization of the muscle’s hot spot regardless of neuronavigation use, the key factor explaining the differences in efficacy was likely the more stable coil positioning with navigation.

In repeated sessions, neuronavigation also ensures reproducibility of coil placement. Without navigation, target localization typically relies on markers (e.g., on a cap), inevitably leading to shifts in coil position and orientation during and between sessions. Deviations from the target average approximately 10 mm for non-navigated rTMS versus 0.3 mm for navigated rTMS. Such coil displacement relative to the target results in lower calculated electric field strength in the cortical target area during non-navigated rTMS, in some cases reduced to 50% of the planned intensity. Another clinically relevant aspect is the significant variability in coil positioning accuracy when non-navigated rTMS is performed by different operators, making the procedure operator-dependent [56].

Structural MRI-based neuronavigation

Structural MRI-based neuronavigation allows for precise localization of the target cortical area and high-spatial-accuracy positioning of the coil. It is particularly valuable for stimulating targets difficult to identify using superficial landmarks. Examples include the auditory cortex, Broca’s area homologue, or the angular gyrus. For instance, structural MRI enables clear localization of pars triangularis and pars opercularis within the inferior frontal gyrus, where rTMS exerts differential effects on speech fluency and object naming in patients with aphasia [57]. Structural MRI navigation is also highly significant for selecting rTMS targets based on specific lesion localization. For example, in patients with drug-resistant focal epilepsy, navigated rTMS of the epileptic activity focus demonstrated a statistically significant seizure frequency reduction (from an average of 8.9 to 1.8 per week). This protocol, when using a figure-8 coil, inherently requires mandatory neuronavigation [58].

Convincing evidence of significant improvement in rTMS efficacy with structural neuronavigation has not yet been obtained, and studies directly comparing the effectiveness of the same rTMS protocol with and without neuronavigation are extremely scarce. In a randomized study by P.B. Fitzgerald et al., a trend toward a more pronounced antidepressant effect of rTMS with structural MRI neuronavigation was observed compared to the standard 5 cm rule targeting approach [28, 59]. However, another study in depression found no statistically significant increase in the efficacy of iTBS to the left DLPFC when using neuronavigation [60]. For tinnitus, direct comparison of structural MRI-navigated and non-navigated rTMS also revealed no statistically significant differences in effect size [61, 62]. In patients with neuropathic pain, no significant increase in analgesic effect was established with neuronavigation, though prolongation of the effect was demonstrated — significant efficacy persisted only in the neuronavigation subgroup one week after completing the rTMS course [63].

It should be noted that some effective rTMS protocols were initially proposed using neuronavigation, such as the TMS-Cog protocol, which is the only protocol with level C evidence of efficacy for Alzheimer’s disease and involves navigated rTMS of 6 cortical areas combined with cognitive training [64, 65]. Another example is the aforementioned protocol of low-frequency rTMS to cortical areas corresponding to the epileptic focus, where non-navigated rTMS is fundamentally impossible [58].

Functional MRI-based neuronavigation

Even within a single anatomical region, since its dimensions typically exceed the focal area of the magnetic field generated by a given coil type, numerous target localization options and consequently stimulating coil positions are possible. The most striking example is the DLPFC, which is large and functionally highly heterogeneous. In fact, even with structural MRI-based neuronavigation, DLPFC target localization is determined arbitrarily [18]. For instance, in the aforementioned study by P.B. Fitzgerald et al., the rTMS target within the DLPFC was standardized at the junction of Brodmann areas 46 and 9 [59], despite lacking direct compelling evidence supporting this specific approach. Similarly, the selection of M1 regions for rTMS protocols treating conditions such as pain, PD, insomnia, migraine, bladder and bowel dysfunction, and chronic disorders of consciousness is arbitrary, as these cases clearly require targeting areas with distinct connectivity patterns to other brain regions [16, 37, 38, 66]. Thus, when navigating solely via structural MRI, precise coil positioning is feasible but it is unclear which specific cortical area should be selected. Stimulated areas may exhibit varying cytoarchitectonic organization and structural connectivity, characterized by high interindividual variability, particularly critical in the context of rTMS network effects [18, 31, 67].

fMRI techniques allow for various methods to assess the functionality of a given brain area and associate it with the modulated function. In particular, fMRI with a behavioral paradigm to select stimulation targets is generally based on the following rationale: if activation is observed in a specific area of the cerebral cortex during the performance of a paradigm (task), then that area is significant for the specific process (function); accordingly, its stimulation will affect that process (function). Following this logic, the paradigm must correspond to the function whose modulation is the objective of conducting rTMS [68, 69]. A necessary caveat when using this approach is understanding the correlative nature of activation observed in fMRI data and behavioral phenomena; therefore, fMRI with a paradigm is typically used to refine the specific individual localization of activation within a broad cortical area, whose role in supporting the studied process (function) has been confirmed using various other methods.

For instance, fMRI with cognitive paradigms is widely used in studies on modulation of cognitive functions, particularly working memory, in healthy volunteers. Several paradigms have been described, such as the n-back test [68] and modified Sternberg task [69], for personalized targeting of rTMS within the DLPFC. A similar approach can be used for target identification when applying rTMS to treat cognitive impairments. For example, D.Yu. Lagoda et al. demonstrated the efficacy of high-frequency rTMS of the left DLPFC in patients with vascular cognitive impairment, using as the target a region within the DLPFC showing maximal activation during a serial counting task paradigm, which can be considered a paradigm assessing executive brain functions [70].

fMRI with a task-based paradigm is promising in patients with structural lesions involving reorganisation of functionally significant motor or speech areas. Although protocols inhibiting activity in homologous contralateral hemisphere regions are currently predominantly used for these disorders [7], pathophysiologically, protocols with an LTP-like effect on preserved cortical areas in the affected hemisphere appear more justified. In the absence of neuronavigation for patients with aphasia or post-stroke hemiparesis, when no MEP is elicited in cortical lesions, adequate target selection for rTMS is impossible, as coil placement in standard positions often results in stimulation of nonviable tissue. Speech or motor paradigms allow localization of preserved functionally significant areas for subsequent personalized stimulation [71–73]. A similar approach has been described for target selection based on activation areas in the occipital cortex during visual paradigm tasks in patients with post-stroke hemianopia [74]. fMRI or other brain mapping methods for target determination are also relevant in patients where MEP cannot be obtained despite intact primary motor cortex (e.g., in spinal cord injury and phantom pain syndrome after amputation) [75].

When using fMRI activation maps to select targets in patients with mental disorders, achieving a perfect match between the paradigm and the modulated function is impossible. However, several studies have proposed specific paradigms that partially correspond to disease symptoms. For example, in generalized anxiety disorder, an approach has been described for identifying rTMS targets using activation maps during a computer-based gambling paradigm involving uncertainty and frustration [76]. For verbal hallucinosis, a method for target localization in the auditory cortex based on activation during a speech recognition paradigm has been proposed [77]. In patients with treatment-resistant depressive episodes, particularly those with cognitive impairments, cognitive paradigms such as the Stroop test or face/shape recognition and matching tasks may be used to identify targets [78].

fMRI with paradigms is also promising for target selection in M1 during rTMS for patients with disorders unrelated to corticospinal tract involvement. For instance, a series of studies by researchers from Shanghai University demonstrated that in healthy individuals, the hot spot area and peak activation during fMRI with a motor paradigm (finger tapping) are significantly spatially distinct and exhibit different functional connectivity patterns with other motor and non-motor structures [79]. Here, the area of maximal activation shows stronger connectivity with subcortical structures (putamen, globus pallidus), making it appear more suitable as a therapeutic TMS target for PD [38]. A recent clinical case reported using the activation area during a “foot tapping” paradigm as an rTMS target in a PD patient [80]. More precise target definition within the M1 is highly promising for rTMS in this region when treating neuropathic pain [37].

A key limitation of the task-based fMRI paradigm method is the requirement for active participation of the patient or healthy volunteer in task performance and the dependence of results on their engagement level. Consequently, the method is inapplicable when task performance is impossible due to severe functional impairment. Additionally, the low signal-to-noise ratio in calculating individual activation maps results in poor reproducibility and reliability of the obtained targets (see details in [81]).

Resting-state functional MRI-based navigation

Application of resting-state fMRI data for defining rTMS targets currently represents the most actively researched direction in personalized target selection. The core idea involves shifting toward the concept of network genesis for a number of pathological conditions and the possibility of modulating activity not of a specific area, but of the entire network via rTMS [82]. Within this concept, a pathological condition is explained by disruptions in interactions within a specific network or between networks, which can be identified through resting-state fMRI analysis and subsequently used as targets for neuromodulation aimed at restoring impaired interactions [83–86].

The impetus for developing this direction came from data obtained by M. Fox et al., demonstrating a significant negative correlation between the effect of rTMS and the group-averaged functional connectivity between the target and the subgenual cingulate cortex in treatment-resistant depression [87]. These findings were later confirmed in other cohorts [28, 88]. Based on this, it was proposed to define areas in the left DLPFC most anti-correlated with the subgenual cingulate cortex as targets for rTMS. Several small studies using this approach yielded heterogeneous results; however, overall, no significant increase in the antidepressant effect of the personalized protocol compared to standard protocols was found [18, 89, 90].

The Stanford Neuromodulation Therapy deserves special mention. It is based on the aforementioned approach of personalized target selection for rTMS using resting-state fMRI and involves administering 10 triple iTBS protocols a day at 1-hour intervals over 5 days. A randomized study demonstrated significant antidepressant efficacy of this protocol compared to sham control in terms of remission rate (57% vs. 0%, respectively) and response rate (71.4% vs. 13.3%, respectively) [91]. In 2022, the Stanford Neuromodulation Therapy, including its algorithm for individualized target determination based on functional connectivity with the subgenual cingulate cortex, was approved by the FDA. Subsequent data confirming the antidepressant effect of this protocol were reported in several smaller studies [92–94].

Personalized rTMS target determination using resting-state fMRI functional connectivity analysis is also being investigated for numerous other disorders. For example, target selection based on functional connectivity with the amygdala has been proposed for patients with post-traumatic stress disorder [95] and addictions [96].

The advancement of this field is largely associated with improvements in resting-state fMRI data analysis algorithms. To improve the signal-to-noise ratio and enhance signals from small regions (e.g., the subgenual cingulate cortex area), additional algorithms have been proposed, such as the “seed map” algorithm, where the signal for calculating individual connectivity is obtained not by averaging over a small region but by weighted averaging across the entire brain with weights equal to the group connectivity of that region [97]. Furthermore, the selection of the point whose coordinates will subsequently be chosen as the target can be performed using different methods [28]. In the simplest “classical” approach, the point with the maximum value of the functional connectivity characteristic of interest is selected (e.g., in the context of subgenual cingulate cortex DLPFC interaction — the most anti-correlated voxel) [98]. However, in cases of high signal heterogeneity in the studied region, this approach may yield false, poorly reproducible results. The other two approaches account for signals from the surrounding area, proposing either to use the point with the maximum average signal within a sphere of a specified radius around it (searchlight algorithm) [97] or to initially threshold the cluster with the maximum signal using a set cutoff and employ the center-of-gravity of this cluster (cluster-based method) [98]. In these cases, the results depend on several parameters predefined by researchers: for the searchlight algorithm — the sphere radius; for the cluster-based algorithm — the cluster threshold. An example of personalized target localization is presented in Fig. 3.

 

Fig. 3. An example of target localization in the left DLPFC and the left posterior parietal cortex based on individual functional connectivity analysis in a healthy volunteer (in-house data).

 

Another possible approach to modulating individual resting-state networks using rTMS is to select a hub of a specific neuronal network as the target, for example, the default mode network or the frontoparietal control network. This method of target selection for rTMS has been described, in particular, for modulating cognitive functions. Several algorithms have been proposed for identifying individual resting-state networks.

In a study by D. Wang et al., an iterative method for constructing individual networks was developed [99]. A resting-state network atlas obtained from a group of healthy subjects was used as the initial approximation. Subject-specific mean BOLD signals from these networks were taken as reference signals, and their correlation with signals from all cortical surface vertices defined the sets of regions corresponding to these networks at each iteration step. Signals averaged across these region sets, combined with reference signals, were assigned as new reference signals, and the procedure for refining the areas was repeated until significant changes ceased.

Another algorithm was proposed by E.M. Gordon et al. and is based on applying a threshold to the individual correlation matrix of all vertices on the cortical surface [100]. The Infomap algorithm was applied to the resulting binarized connectivity matrix to identify communities representing resting-state networks. As in all methods that do not initially rely on a group atlas, this algorithm requires a matching procedure to assign well-known names to the identified networks, which in this case is based on maximizing the Jaccard coefficient between individual and group networks.

Another approach was proposed by R. Kong et al. and is termed the multi-session hierarchical Bayesian model [101]. A model with parameters was constructed, describing the connectivity profiles of resting-state networks using their group mean values and different types of variability (between subjects, within a subject between sessions, and between different regions of the same network). The parameters of this model were estimated from a group of healthy volunteers and then applied to construct resting-state networks for individual subjects. The independent component analysis method is also used to construct individual resting-state networks. In this case, to assign commonly accepted names to the identified components, they must be matched with group data, which can be done visually [102] or algorithmically [103, 104]. The main limitation of visual matching is the high dependence of the results on the subjective opinion of the expert. Algorithmic matching is free of such subjectivity but also has limitations, especially in borderline cases when a given set of regions is approximately equally similar to several known networks.

The role of spatial navigation in determining coil orientation and rTMS intensity

In recent years, it has become evident that spatial navigation in rTMS is important not only for target selection and coil positioning but also for determining correct coil orientation and stimulation intensity. Current understanding of TMS biophysics and physiology indicates that regardless of target selection method, precise identification of the stimulated cortical area is crucial for determining appropriate coil orientation and stimulation intensity [105–107]. During TMS, axons are primarily excited rather than neuronal cell bodies, with axon depolarization probability dependent on their spatial orientation relative to the induced electric field [107]. Studies show that the maximum induced field in a gyrus occurs when current direction is perpendicular to the gyrus axis [106]. For example, during motor cortex TMS, the highest amplitude MEPs are recorded when the coil is oriented at 45° to the sagittal plane, as this positions the induced electric field perpendicular to the precentral gyrus. Altering coil orientation without displacement significantly affects MEP amplitude, and certain orientations may yield no detectable MEPs [108].

For non-motor cortical regions, due to the extremely high variability in the shape and topography of gyri, the required coil orientation could only be determined via neuronavigation [109, 110]. Different coil orientations alter the location of the area of predominant neuronal activation. When the coil is not oriented perpendicularly to the underlying gyrus, stronger neuronal activation may occur at some distance from the coil center, in a cortical area that is more perpendicularly oriented relative to the coil [109]. The displacement of the predominant neuronal activation area depending on coil orientation can reach several centimeters, averaging about 1.6 cm [105]. During rTMS of the DLPFC, changing the coil orientation without altering its position can lead to predominant activation of hubs in different neuronal networks, such as the default mode network and the frontoparietal network [105, 109].

The discussed issue is closely related to determining TMS intensity. Even for rTMS of non-motor areas, intensity is typically calculated based on motor threshold as an accessible indicator of cortical excitability, which lacks physiological justification. Moreover, interindividual differences in motor threshold are primarily associated not with the intrinsic excitability of the motor cortex, but with the distance from the coil to the cortex and the individual orientation of corticospinal fibers relative to the coil [111]. A possible alternative is calculating the electric field strength in the target cortical area, achievable through modeling of varying complexity in several neuronavigation systems. This approach, generally termed electric field navigation, accounts for the distance from the coil to the target cortical region with attention to tissue conductivity and the orientation of the electric field relative to the target cortical area [19, 109, 112, 113].

Equally important but far less studied is the relationship between the physiological effects of TMS and coil orientation, considering both macro- and microanatomy of the cortex. Studies on motor cortex TMS demonstrate that changing coil orientation and, consequently, the direction of induced current in the cortex affects the pool of activated neurons [114]. Therefore, optimal coil orientation should depend on the target neuronal population. For instance, the conventional coil orientation for motor cortex stimulation adequate for direct and transynaptic excitation of layer V pyramidal cells might not be such for rTMS of the motor cortex in neuropathic pain, as the micro-level target for TMS in this case is likely projection fibers traversing the superficial layers of the motor cortex, which have a different orientation [37].

Conclusion

This review demonstrates the evolution of approaches to rTMS target identification that has occurred over the past 5–10 years. The standard approach, which involves coil placement based on external landmarks, often fails to reach the target cortical area and does not account for individual variations in cortical anatomy or functional area locations. Neuronavigation allows precise coil positioning with high spatial resolution relative to the selected cortical region and enables real-time position monitoring. Moreover, the capabilities of navigational control have spurred the development of an entirely new research field focused on identifying personalized targets within anatomical regions using structural and functional neuroimaging methods.

At the same time, the current landscape of rTMS target identification approaches is ambiguous. Nearly the entire evidence base for therapeutic rTMS, which underpins clinical guidelines and practical application, is derived from studies employing standard target identification and coil positioning methods. To date, there is a lack of convincing empirical evidence demonstrating significant improvement in therapeutic rTMS efficacy when using neuronavigation and various modern approaches for precision personalized target identification. Undoubtedly, in routine clinical practice, standard target identification approaches which achieve clinical effects of established magnitude and variability for many conditions, as demonstrated in relevant clinical trials, can and should predominantly be used.

On the other hand, randomized controlled trials directly comparing clinical effects between protocols using different target localization methods remain scarce and have been conducted on small cohorts, allowing detection of only strong effects. The development of new approaches for rTMS target selection inevitably encounters numerous methodological issues and challenges concerning the choice of specific neuroimaging modalities for target identification, multiple issues in analyzing complex neuroimaging data, their reproducibility, etc. Essentially, current efforts involve testing precision targeting methodologies and identifying potentially most effective and reproducible approaches that should subsequently be investigated in clinical studies.

Significant advances in understanding rTMS biophysics and physiology demonstrate its effects’ extreme dependence on both preceding and ongoing neuronal activity, as well as numerous protocol parameters [31]. In this context, achieving strong and reproducible rTMS effects necessitates personalized adaptive state-dependent multifocal neuromodulation of neuronal networks. Progress in this direction is impossible without neuronavigation, which enables precise coil positioning, correct orientation relative to cortical targets, and selection of appropriate stimulation intensity. However, it is evident that successful translation of these advances into clinical practice critically requires resolving methodological issues and conducting high-quality clinical trials.

About the authors

Ilya S. Bakulin

Russian Center of Neurology and Neurosciences

Author for correspondence.
Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0003-0716-3737

Cand. Sci. (Med.), senior researcher, Head, Noninvasive neuromodulation group, Institute of Neurorehabilitation

Russian Federation, Moscow

Alexandra G. Poydasheva

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0003-1841-1177

Cand. Sci. (Med.), researcher, Noninvasive neuromodulation group, Institute of Neurorehabilitation

Russian Federation, Moscow

Dmitry O. Sinitsyn

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0001-9951-9803

Cand. Sci. (Physics and Mathematics), senior researcher, Noninvasive neuromodulation group, Institute of Neurorehabilitation

Russian Federation, Moscow

Anastasia N. Sergeeva

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0002-2481-4565

Cand. Sci. (Med.), researcher, Radiology department, Institute of Clinical and Preventive Neurology

Russian Federation, Moscow

Alfiia Kh. Zabirova

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0001-8544-3107

junior researcher, Noninvasive neuromodulation group, Institute of Neurorehabilitation

Russian Federation, Moscow

Dmitry Yu. Lagoda

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0002-9267-8315

Cand. Sci. (Med.), researcher, Noninvasive neuromodulation group, Institute of Neurorehabilitation

Russian Federation, Moscow

Dmitry V. Sergeev

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0002-9130-1292

Cand. Sci. (Med.), senior researcher, Department of consciousness and memory, Institute of Neurorehabilitation, scientific secretary

Russian Federation, Moscow

Natalia A. Suponeva

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0003-3956-6362

Dr. Sci. (Med.), Corr. member of RAS, Director, Institute of Neurorehabilitation

Russian Federation, Moscow

Michael A. Piradov

Russian Center of Neurology and Neurosciences

Email: annaly-nevrologii@neurology.ru
ORCID iD: 0000-0002-6338-0392

Dr. Sci. (Med.), Professor, Full member of RAS, Director, Vice-President of the Russian Academy of Sciences

Russian Federation, Moscow

References

Supplementary files