Towards optimizing repetitive transcranial magnetic stimulation protocols for chronic subjective tinnitus

1 Introduction

Chronic subjective tinnitus is the sense of sound in the absence of an external acoustic source (Tyler, 2000). It affects about 10%−15% of individuals and is sometimes associated with sleep disruption, diminished focus, anxiety, sadness and decreased quality of life (Kleinjung et al., 2024). The clinical load is not just dictated by the loudness of the heard sound but also by the attention, salience, threat value and emotional meaning attributed to the phantom sound.

1.1 Why tinnitus occurs: peripheral deafferentation and central gain

Tinnitus usually starts with a reduction or distortion of auditory input due to noise exposure, age-related hearing loss, ototoxicity, cochlear injury, or synaptopathy (Liberman and Dodds, 1984). Neuronal gain may increase to compensate for reduced afferent input in the central auditory system. This response is initially adaptive, but can become maladaptive, leading to decreased inhibition, increased spontaneous activity, abnormal spike timing, and increased neuronal synchrony in the dorsal cochlear nucleus, inferior colliculus, medial geniculate body, and auditory cortex (Auerbach et al., 2014). This internally produced activity of the brain system might therefore be perceived as sound.

There is experimental evidence supporting this essential contribution. Salicylate-induced tinnitus demonstrates that reversible cochlear malfunction can result in central auditory hyperactivity without lasting structural damage (Chen et al., 2013). Besides auditory nuclei, trigeminal and cervical somatosensory inputs may also impact auditory nuclei after hearing loss, which might contribute to explaining somatic modulation and interindividual variability (Ralli et al., 2017).

1.2 Thalamocortical dysrhythmia and impaired gating

Thalamocortical dysrhythmia provides a way for aberrant auditory activity to attain conscious perceptual prominence. Auditory deafferentation causes hyperpolarization of thalamic relay neurons, especially in the medial geniculate body, and shifts activity from tonic transmission to pathological low-frequency bursts. Increased theta activity and decreased alpha-mediated inhibition reduce sensory gating. Gamma activity at the borders of dysrhythmic regions, the proposed “edge effect,” may enhance perceptual binding and conscious awareness (Llinás et al., 1999; Knipper et al., 2013; De Ridder et al., 2015; Brinkmann et al., 2021). This suggests that abnormal theta–gamma coupling converts dysregulated thalamocortical activity into a persistent phantom percept.

1.3 Distributed auditory and non-auditory networks

Persistent tinnitus is not only an auditory cortical disorder. With chronicity, maladaptive activity recruits dorsolateral prefrontal cortex (DLPFC), anterior cingulate cortex (ACC), insula, temporoparietal junction (TPJ), limbic regions, salience networks, memory systems and somatosensory pathways (Vanneste et al., 2010; De Ridder et al., 2011, 2014; Shahsavarani et al., 2021). Impaired limbic–auditory gating may inhibit habituation, whereas prefrontal and cingulate dysregulation may promote attentional capture, negative evaluation, and distress. Tinnitus severity may thus not correlate with loudness: one individual may dismiss equal sensory activity, while another may find it invasive and frightening. This model is self-reinforcing: peripheral deafferentation induces increased central gain, decreased inhibition and synchrony consolidates abnormal auditory activity, thalamocortical dysrhythmia allows conscious access, and limbic, attentional and salience systems maintain distress and attentional priority (Llinás et al., 1999; Chambers et al., 2016; Saeed and Khan, 2021; Jiang et al., 2025; Figure 1).

1.4 Neurophysiological basis of rTMS

Transcranial magnetic stimulation employs a rapidly changing magnetic field to create an electric current in the superficial cortex. When applied to the motor cortex, the descending corticospinal volley triggered by stimulation produces a motor-evoked potential and allows for the calculation of the motor threshold, which is often used to individualize treatment intensity. Repetitive stimulation can have long-lasting effects by modulating synaptic effectiveness, cortical excitability and network connectivity (Eldaief et al., 2022; Siebner et al., 2022).

Low-frequency rTMS ( ≤ 1 Hz) is commonly linked to inhibitory or long-term depression-like modulation and has been utilized to decrease tinnitus-related auditory hyperactivity. High-frequency rTMS, usually >5 Hz, can increase cortical activity or induce a momentary “virtual lesion” by interrupting current disease processing. High-frequency stimulation is designed to improve top-down attentional and affective control when applied to the DLPFC, rather than directly inhibit auditory activity. These effects are not absolute, but depend on cortical state, target, coil direction, intensity, dosage and individual neurobiology (Fregni and Pascual-Leone, 2007; Weisz et al., 2012; Kricheldorff et al., 2022). Theta-burst and paired-associative paradigms further exemplify the capacity of the temporal structure of stimulation to modulate plasticity (Huang et al., 2024; Nätkynmäki et al., 2024; Figure 2).

1.5 Clinical position, protocol heterogeneity, and research gap

rTMS is interesting because of its ability to directly modify dysfunctional cortical and network activity without surgery. However, the off-label use of rTMS for tinnitus is still exploratory and established tinnitus therapy continues to focus on counseling, sound-based therapies, hearing rehabilitation if required, and cognitive-behavioral interventions. Trials and meta-analyses provide evidence of probable benefit in selected patients, but do not establish rTMS as a universally effective or superior therapy (Liang et al., 2020; Lu et al., 2024; He et al., 2025).

The lack of standardized stimulation protocols remains a major barrier to clinical translation. Target, frequency, intensity, pulses per session, cumulative dosage, session number, target sequence, localization, sham credibility, outcome measure, responder definition, follow-up and patient phenotype (Schoisswohl et al., 2019) vary. These distinctions influence whether the stimulation mainly impacts auditory hyperactivity, prefrontal cognitive-emotional control or a larger tinnitus network. It is necessary to have a sham design since the THI, TQ, TFI, and visual analog ratings are affected by expectation and reporting effects. A credible sham condition should reproduce coil noise, scalp vibration, pressure and facial or scalp muscle sensations without generating a biologically meaningful intracranial electric field. Differences in blinding believability and residual stimulation are seen for coil angulation, sound-only simulation, sensory matching and specially shielded sham coils (Duecker and Sack, 2015; Rossi et al., 2021). Weak sensory matching may inflate apparent efficacy via unblinding, and an active residual field may minimize active-sham differences.

Accordingly, this review aimed to synthesize clinical rTMS protocols for chronic subjective tinnitus and identify how frequency, target, intensity, pulse dose, treatment duration, localization, sham design, and participant characteristics contribute to the observed heterogeneity (Figure 3).

2 Methods

2.1 Review design and eligibility

The review was structured in compliance with PRISMA 2020 (Page et al., 2021). Eligibility was determined according to Population, Intervention, Comparator, Outcomes, Timing and Study design (PICOTS) criteria (Table 1). The main review population comprised adults with persistent subjective non-pulsatile tinnitus. Eligible studies were clinical rTMS trials with low- or high-frequency, combined-frequency or multisite procedures. Sham, active control, alternate rTMS procedures and other pre-specified comparator conditions were eligible. Eligible studies provided a validated tinnitus outcome such as THI, TQ, TFI or a VAS for loudness or distress. Basic mechanistic and technical investigations were utilized to interpret the creation of protocols, with these studies being segregated from the main clinical synthesis.

Table 1. Eligibility criteria.

2.2 Information sources and search strategy

A systematic literature search was performed in PubMed/MEDLINE, Embase, Cochrane Library, principally the Cochrane Central Register Of Controlled Trials (CENTRAL), and Web of Science using controlled vocabulary and free-text words related to chronic subjective tinnitus and repetitive transcranial magnetic stimulation. Search terms included “tinnitus,” “chronic subjective tinnitus,” “subjective tinnitus,” “persistent tinnitus”, “repetitive transcranial magnetic stimulation,” “rTMS,” “transcranial magnetic stimulation,” “auditory cortex,” “temporoparietal junction,” “dorsolateral prefrontal cortex,” and “neuromodulation.” The search was conducted iteratively with single phrases and clinically appropriate combinations of terms according to the search capabilities available inside each database rather than a single prespecified Boolean search string. Search results across the multiple keywords and databases were combined, duplicate entries were deleted, and possibly relevant records were analyzed according to the prespecified PICOTS eligibility criteria. Reference lists of pertinent clinical studies and reviews were screened for possibly eligible papers not returned in the database searches. To increase repeatability, the database utilized, search terminology input, relevant search fields or controlled-vocabulary words, search constraints, date of each search and number of records retrieved are noted and presented in Supplementary Table S1 for each search iteration. The review protocol was not registered prospectively in PROSPERO or any other systematic-review registry. The review was performed according to prespecified PICOTS eligibility criteria and reported following the PRISMA 2020 reporting structure.

2.3 Study selection and data extraction

Titles and abstracts were screened, complete texts checked against pre-specified criteria for eligibility, and duplicates removed. Records were removed if they were not related to chronic tinnitus, were based on animal or non-clinical designs, did not contain a rTMS intervention arm, did not describe acceptable stimulation settings, or offered no validated tinnitus outcome. The study-selection procedure was performed by successive multi-author review. Author 1 conducted the literature search and selection of possibly relevant records. Author 2 then examined the selected studies against the prespecified eligibility criteria and undertook the first screening of titles, abstracts and possibly suitable full-text publications. Authors 3 and 4 independently examined eligibility criteria and critically reappraised the appropriateness of candidate studies for inclusion. Disagreements were resolved through discussion, and unresolved inclusion decisions were finalized by Author 5 following consensus among the authors. The extracted variables were author and year, design, sample size, participant characteristics, cortical target, frequency, intensity (relative to motor threshold), pulses per session, number and duration of sessions, coil and localization method, sham or comparator, follow-up, outcome measures, responder definition, adverse events, and primary findings. Extracted data and eligibility for studies were then discussed within the author team to ensure compatibility with the pre-specified PICOTS framework and the aims of the systematic review.

2.4 Risk-of-bias assessment procedures

Risk of bias was formally assessed using study-design-appropriate validated tools. Randomized parallel-group trials were evaluated using the Cochrane Risk of Bias 2 (RoB 2)tool (Sterne et al., 2019), while randomized crossover trials were assessed using the corresponding RoB 2 framework for crossover designs. Non-randomized comparative intervention studies were evaluated using ROBINS-I, whereas quasi-experimental and other eligible non-randomized clinical designs were assessed using the JBI Critical Appraisal Checklist for Quasi-Experimental Studies, as appropriate to the confirmed study design. For randomized trials, the assessment considered bias arising from the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. For non-randomized studies, additional design-specific sources of bias, including confounding, participant selection, and intervention classification, were considered where applicable. Authors 1 and 2 performed the initial domain-level risk-of-bias assessment. Authors 3 and 4 independently verified the judgments against the original study reports and relevant methodological information. Disagreements were resolved through discussion, and unresolved assessments were finalized by Author 5 following consensus among the authors. Risk-of-Bias judgements were used to inform the qualitative weighting and interpretation of the clinical evidence rather than as post hoc exclusion criteria. Contextual reports that were not part of the 28- study primary qualitative synthesis were excluded from the formal risk-of-bias assessment. Study-design classification and corresponding risk-of-bias tool allocation are summarized in Supplementary Table S2. Detailed domain-level assessments are provided in Supplementary Tables S3, S4, and the assessment workflow is presented in Supplementary Figure S1.

2.5 Evidence classification and qualitative synthesis

Studies were classified as: (1) low-frequency protocols, mainly 1 Hz stimulation of auditory, temporal, or temporoparietal regions; (2) high-frequency protocols, generally ≥10 Hz and often aimed at transient suppression or excitatory modulation; or (3) combined-frequency/multisite protocols stimulating multiple frequencies or network nodes, often combining prefrontal with auditory or temporoparietal stimulation. Methodological quality was qualitatively considered on the basis of study design, randomization, blinding, sham credibility, completeness of protocol reporting, attrition, outcome selection, and duration of follow-up. Risk-of-bias evaluations were incorporated into the qualitative interpretation of each study, rather than being used as exclusion criteria or quantitative weights. Findings from studies with some concerns or high/serious methodological constraints were viewed with greater caution and required verification from stronger evidence, whereas studies with higher methodological credibility were interpreted with greater confidence. The pre-specified rules to govern the weighing of qualitative evidence are detailed in Supplementary Table S5.

The studies were substantially different in population, target, frequency, intensity, dose, number of sessions, localization, sham design, outcome measure and follow-up. For this reason, meta-analysis was not performed. Qualitative synthesis therefore focused on the direction, durability and methodological context of findings rather than pooled effect estimation.

3 Results

3.1 Study selection and evidence profile

The search identified 210 records: 200 from database searching and 10 from other sources. Duplicates (n = 160) were removed, and titles and abstracts of 50 records were screened. Ten records were excluded. All 40 reports requested for retrieval were obtained and reviewed in full text. Twelve reports were excluded due to reasons such as lack of methodological detail, unclear or missing stimulation parameters, lack of validated tinnitus outcomes, or lack of adequate clinical data, leaving 28 studies in the primary qualitative synthesis (Figure 4). Two additional contextual reports-a clinical case report and an emerging digital twin application-were considered separately in the discussion/future directions and were not included in the 28-study primary synthesis, primary evidence matrices, or formal risk-of-bias assessment.

The evidence ranged from randomized sham-controlled trials, double-blind crossovers, active-comparator trials, pilot studies, observational investigations and early proof-of-concept experiments. Target selection, laterality, frequency, motor-threshold intensity, dose of pulses, number of sessions, localization, control design, responder definition, outcome and follow-up were heterogeneous. The principle characteristics of the participants, stimulation protocols, follow-up periods, and clinical findings of the 28 included studies are summarized in Table 2.

Table 2. Concise clinical summary of the 28 studies included in the primary qualitative synthesis.

AC, auditory cortex; DLPFC, dorsolateral prefrontal cortex; MT, motor threshold; NR, not reported; RMT, resting motor threshold; STG, superior temporal gyrus; THI, Tinnitus Handicap Inventory; TQ, Tinnitus Questionnaire; TPJ, temporoparietal junction; VAS, visual analog scale.

3.2 Risk of-bias profile across included studies

Risk of bias varied across the 28 included studies. Overall, 12 studies (42.9%) were classified as having some concerns, 4 (14.3%) as high risk, and 12 (42.9%) as serious or higher risk; no study was rated overall low risk. Among randomized studies assessed using RoB 2, the primary issues were poor reporting of allocation concealment, difficulties in blinding or sham credibility, missing outcome data, and uncertainty regarding selective reporting. Non-randomized and quasi-experimental studies had greater methodological limitations, particularly confounding, participant selection, lack of suitable contemporaneous controls and use of subjective outcomes. Risk-of-bias judgments were therefore used to qualify the strength of inference rather than to exclude otherwise eligible studies. Study-design classification and corresponding risk-of-bias tool allocation are summarized in Supplementary Table S2 detailed domain level assessments are provided in Supplementary Tables S3, S4 and the assessment workflow is presented in Supplementary Figure S1.

3.3 Low-frequency rTMS

The predominant approach was low-frequency rTMS, most often 1 Hz over the left auditory cortex, superior temporal gyrus or TPJ. Early case series and small neuronavigated studies reported reductions in tinnitus severity, sometimes for several months (Eichhammer et al., 2003; Kleinjung et al., 2005, 2007; Langguth et al., 2006; Marcondes et al., 2010). Sometimes the response was larger in patients with shorter tinnitus duration, preserved hearing, or greater baseline burden.

The amount of controlled evidence varied. Folmer et al. (2015) found that active bilateral temporal stimulation had a higher responder rate than sham (56% versus 22%) and that the benefit lasted up to 26 weeks. By contrast, Hoekstra et al. (2013), Roland et al. (2016), Sahlsten et al. (2017) and the large multicenter trial by Landgrebe et al. (2017) failed to demonstrate consistent superiority over sham. Delayed improvement at later follow-up was reported by Anders et al. (2010) and no clear advantage of MRI navigation over 10–20 EEG positioning was found by Sahlsten et al. (2019). These results indicate that repeated 1 Hz auditory stimulation may be beneficial for selected participants, but it is not universally effective nor reliably explained by anatomical accuracy alone. The stimulation characteristics and principal findings of the low-frequency rTMS studies are summarized in Table 3.

Table 3. Low-frequency rTMS studies.

AC, auditory cortex; CBT, cognitive-behavioral therapy; DLPFC, dorsolateral prefrontal cortex; MT, motor threshold; RMT, resting motor threshold; STG, superior temporal gyrus; TPJ, temporoparietal junction.

3.4 High-frequency rTMS

High-frequency temporal stimulation has been mainly investigated in small single-session or crossover experiments. Plewnia et al. (2003) identified left temporoparietal sites where 10 Hz trains produced a transient suppression of tinnitus, consistent with a location-specific “virtual lesion” effect. Folmer et al. (2006) reported partial suppression in 6 of 15 participants with duration ranging from minutes to several days. Fregni et al. (2006) showed that 10 Hz left temporoparietal rTMS reduced tinnitus in 3 out of 7 participants but the effect lasted less than 5 min. The high-frequency/proof-of-concept matrix retained PET-guided low-frequency dose testing by Plewnia et al. (2007), because it directly informed brief experimental paradigms, showing dose-dependent suppression lasting up to around 30 min. In conclusion, these small studies support transient modulation but not durable therapeutic efficacy. The protocols and principal findings of the high-frequency rTMS studies are summarized in Table 4.

Table 4. High-frequency rTMS studies.

BA, Brodmann area; MT, motor threshold; PET, positron emission tomography; RMT, resting motor threshold; tDCS, transcranial direct-current stimulation.

3.5 Combined-frequency and multisite rTMS

Multifrequency and multisite methods were developed to engage both the sensory-perceptual and cognitive-emotional components of tinnitus. De Ridder et al. (2005) showed that the frequency corresponding to the maximum short-term suppression was dependent on the duration of tinnitus. However, Khedr et al. (2008) found improvement with 1, 10, and 25 Hz stimulation compared with sham, but no clearly superior frequency. Vanneste et al. (2012) found frequency- and laterality-dependent transient effects of parietal stimulation; De Ridder et al. (2013) implicated right DLPFC and ACC–auditory interactions.

Sequential prefrontal-temporal protocols yielded promising but non-uniform results. Only the combined DLPFC–auditory stimulation showed a lasting benefit after 3 months Kleinjung et al., (2008). In Noh et al. (2019) and Noh et al. (2020), dual auditory–DLPFC stimulation led to a more sustained improvement than sham or auditory-only stimulation. In contrast, Kreuzer et al. (2015) observed comparable gains with ACC → TPJ and DLPFC → TPJ protocols, and Lehner et al. (2016) observed no statistically significant benefit of triple-site over single-site treatment. Multisite stimulation is therefore promising, but superiority is not proven. The combined-frequency, multisite stimulation protocols and their principal findings are summarized in Table 5.

Table 5. Combined-frequency and multisite rTMS studies.

ACC, anterior cingulate cortex; AC, auditory cortex; DLPFC, dorsolateral prefrontal cortex; MT, motor threshold; RMT, resting motor threshold; TPC, temporoparietal cortex; TPJ, temporoparietal junction.

3.6 Target localization and functional guidance

Localization approaches included the international 10–20 EEG system, structural MRI, functional MRI, PET/MRI neuronavigation, electric-field modeling, and tonotopic or pitch-matched targeting. MRI- or PET-guided approaches increased anatomical reproducibility and allowed targeting of individualized hypermetabolic areas. Sahlsten et al. (2019), however, found no major outcome difference between navigated and non-navigated stimulation and several trials of precise targeting were negative. Anatomical accuracy is therefore required for reproducibility, but may not be sufficient if the selected structure is not the dominant functional node in an individual patient.

3.7 Sham credibility and comparator design

The control conditions were quite heterogeneous (Table 6). Coil angulation preserved coil sound but did not fully reproduce somatosensory effects and could produce residual stimulation. Sound-only sham was especially prone to unblinding. Dedicated shielded coils better attenuated cortical field. Quality of sensory matching and blinding assessment were reported inconsistently. Active-comparator studies could compare frequencies, targets or navigation methods, but could not estimate the contribution of expectation independently, repeated clinical contact or spontaneous fluctuation of symptoms. Rossi et al. (2007) found that only about 71% of participants were effectively blinded, indicating that the credibility of the sham must be evaluated, not just assumed.

Table 6. Sham and comparator designs represented in the included evidence.

3.8 Tinnitus outcome domains

The studies reviewed investigated multiple dimensions of tinnitus outcomes, which were interpreted according to the individual concept tested, not as equivalent markers of overall change. The outcome instruments related to tinnitus, domains assessed, assessment time points, and responder definitions used across the included studies are summarized in Table 7. THI was thought to measure tinnitus-related handicap or overall burden, TQ wider tinnitus-related symptom burden, and VAS or numerical ratings of perceptual outcomes such as tinnitus loudness or intensity. Some low-frequency studies showed improvement based on the questionnaire: Khedr et al. (2008) showed THI improvement relative to sham, Marcondes et al. (2010) reported sustained THI improvement, and Kleinjung et al. (2007) reported a ≥5-point TQ reduction in around 40% of participants. Hoekstra et al. (2013) observed no significant active-treatment benefit in either THI or TQ. Kleinjung et al. (2008) observed persistent TQ improvement at 3 months solely with combined prefrontal–auditory stimulation in multisite studies, whereas Noh et al. (2019, 2020) reported more sustained improvement following dual auditory cortex–DLPFC stimulation than with sham or auditory cortex-only stimulation, respectively. In contrast, temporal investigations with high frequency, such as those of (Plewnia et al. (2003), Folmer et al. (2006), and Fregni et al. (2006), indicated mostly temporary suppression of tinnitus perception, rather than lasting improvement in tinnitus-related handicap or functional load. The data suggest that decreases in THI or TQ do not necessarily transfer into reductions in tinnitus loudness and that short-term perceptual suppression does not always translate into persistent improvement in tinnitus-related burden. Thus, outcome variability may contribute to apparent discrepancies between rTMS trials, but comparisons are constrained by variations in responder criteria and follow-up duration.

Table 7. Tinnitus-specific outcome measures used in the included rTMS studies.

THI, Tinnitus Handicap Inventory; TQ, Tinnitus Questionnaire; TFI, Tinnitus Functional Index; VAS, Visual Analog/Analog Scale; VNS, Visual Numerical Scale; NRS, Numeric Rating Scale; TSS, Tinnitus Severity Scale; GIC, Global Impression of Change; rTMS, repetitive transcranial magnetic stimulation.

3.9 Adverse events and tolerability

Safety and tolerability reporting was variable throughout the included trials. Among trials reporting safety results explicitly, rTMS was usually recognized as viable and well-tolerated; however, adverse-event reporting was not standardized. Rossi et al. (2007) reported two withdrawals due to temporary aggravation of tinnitus, while Kreuzer et al. (2015) reported four dropouts including one adverse event and three occurrences of symptom worsening. By contrast, other studies identified rTMS as safe or well-tolerated, including larger controlled and multisite trials. The absence of reported occurrences was not construed as proof that no adverse events occurred, because adverse events were not routinely documented in all investigations. To summarize, the existing data show an overall acceptable tolerability of rTMS in chronic tinnitus. However, poor safety reporting hampers a valid estimate of the frequency and range of treatment-related side effects.

3.10 Overall qualitative synthesis

The direction and duration of rTMS effects were defined by the interaction of target, frequency, cumulative dose, session number, localization, tinnitus duration, hearing status, baseline severity, network state and control credibility. Most extensively studied was repeated 1 Hz auditory or temporoparietal stimulation, but larger sham-controlled trials were often null. Short bursts of high frequency temporal stimulation can transiently disrupt tinnitus perception, but provide little evidence for durable therapeutic plasticity. The best approximation to the distributed-network model was provided by combined auditory-prefrontal approaches, which yielded sustained improvement in selected studies, but did not consistently outperform single-site treatment.

4 Discussion

The aim of this review was to critically synthesize the available clinical evidence on rTMS for chronic subjective tinnitus with attention to stimulation frequency, cortical target, intensity, pulse dose, train structure, number of sessions, target sequence, localization or navigation, comparator validity, outcome selection, and durability of treatment effects. The 28 trials included in the initial qualitative synthesis showed biological activity of rTMS, but clinical effectiveness varied widely, and no particular frequency–target–dose combination was consistently beneficial. Low-frequency stimulation of the auditory cortex or temporoparietal areas varied from prolonged clinical or metabolic benefit in selected studies to no superiority over sham in larger controlled trials. High-frequency temporal stimulation induced largely short-lived perceptual suppression. Sequential auditory–prefrontal procedures in selected experiments produced more persistent effects but did not consistently outperform single-site therapy. Therefore, no evidence is available to claim the general superiority of any particular protocol. The interpretation of these data has to be dissociated from clinical evidence and mechanistic explanation. The direct clinical data summarized in this study show changes in THI, TQ, VAS ratings, responder rates, active-sham differences and durability. Conversely, the clinical patterns presented here may be understood in terms of LTD- or LTP-like plasticity, modulation of central gain, thalamocortical dysrhythmia, oscillatory reorganization, and enhanced prefrontal top-down regulation, all biologically plausible explanations unless directly demonstrated in the respective treatment study. Also, EEG-/MEG-informed targeting, connectivity-guided stimulation and computational modeling are still exploratory techniques awaiting prospective validation. The choice of comparator also determines how apparent treatment efficacy is interpreted. The trials comprised sham coil angulation, sham matched for sound or sensation, specialized sham coils, alternate frequencies or targets, additional neuromodulation approaches, and single-site vs. multisite comparisons. Patient anticipation, placebo response, coil sound, scalp feeling, recurrent clinical interaction and spontaneous fluctuation of tinnitus may all impact patient-reported results. Hence the evidentiary value of an apparent treatment effect depends not just on its size but also on the biological inertness and sensory believability of the comparator.

4.1 Neuroplastic mechanisms in tinnitus pathophysiology: synaptic and network maladaptation

Chronic subjective tinnitus may be seen as a central auditory and distributed network dysfunction, usually triggered but not necessarily perpetuated by peripheral auditory damage. Damage to the cochlea, synaptopathy, loss of hair cells, noise exposure or other kinds of deafferentation lead to a reduction in afferent auditory input and a disturbance in the balance between glutamatergic excitation and GABAergic or glycinergic inhibition. The resultant homeostatic response may promote central gain and disinhibition (Gerken, 1996; Schaette and Kempter, 2006). With chronicity, increased spontaneous firing, abnormal bursting, enhanced neuronal synchrony and altered tonotopic organization may become established within the dorsal cochlear nucleus (DCN), inferior colliculus, medial geniculate body and auditory cortex (Eggermont and Roberts, 2004; Shore et al., 2016). Thalamocortical dysrhythmia, including diminished alpha-mediated inhibition and aberrant low-frequency–gamma coupling, may further contribute to the stabilization and conscious perception of the phantom sound (Llinás et al., 1999). Experiments and computer simulations of the DCN give a mechanistic explanation for how plasticity that is initially compensatory may become harmful. Homeostatic scaling of excitatory and inhibitory synapses may be triggered by reduced auditory-nerve input, and acoustic trauma can change multimodal parallel-fiber synapses on fusiform and inhibitory cartwheel cells. Possible biological substrates for trauma-related metaplasticity and altered subsequent long-term potentiation and depression include increased presynaptic release probability at parallel fiber–fusiform cell synapses (Fujino and Oertel, 2003; Oertel and Young, 2004). Other processes such as NMDA-receptor overactivity, neuroinflammatory signaling involving the tumor necrosis factor-α and interleukins, altered brain-derived neurotrophic factor and changes in endocannabinoid or dopaminergic signaling may contribute to persistent hyperexcitability and maladaptive plasticity, although their direct relevance to clinical rTMS response remains incompletely established (Tan et al., 2005; Wang et al., 2020; Ryan et al., 2018; Shulman et al., 2021). Additional somatosensory–auditory cross-modal plasticity may also account for the fact that in certain individuals tinnitus may be regulated by jaw, head or neck input (Shore et al., 2016). Chronicity also affects tinnitus involvement of the amygdala, hippocampus, DLPFC, anterior cingulate cortex, insula, frontoparietal attention systems, and salience and memory networks. Impaired limbic-auditory gating, negative assessment, and attentional fixation may consequently contribute to protracted awareness and suffering (Rauschecker et al., 2010, 2015). This dispersed design also helps explain why people with a comparable hearing loss or tinnitus loudness may suffer differing impairment dramatically, and why targeted inhibition of the auditory cortex alone may not necessarily yield persistent relief.

4.2 Frequency-dependent neuromodulation: restoring circuit balance via LTD- and LTP-like plasticity

Low-frequency rTMS is generally linked to inhibitory or LTD-like effects, which could plausibly explain the reduction of tinnitus-related auditory hyperactivity, but the included clinical trials did not directly demonstrate LTD as the mediator of symptomatic improvement (Filipović et al., 2010; Peng et al., 2018). Early PET-, SPECT- or MRI-guided investigations have shown clinical improvement linked with lower temporal activity, in some instances lasting many months. Folmer et al. (2015) likewise observed a larger number of responders with active compared to sham stimulation. Conversely, Hoekstra et al. (2013), Roland et al. (2016), Sahlsten et al. (2017) and the large multicenter trial by Landgrebe et al. (2017) have shown little or no active-sham separation. These results suggest that modulation of auditory cortex activity is physiologically possible, but may not be adequate when larger tinnitus-related networks remain maladapted. Temporal stimulation with a high frequency seems to give a distinct clinical picture. The transitory but rapid depression observed by Plewnia et al. (2003), Folmer et al. (2006), Fregni et al. (2006) and De Ridder et al. (2005) is compatible with short-lived interruption or resetting of ongoing cortical activity rather than with established persistent rearrangement. High-frequency stimulation of the DLPFC has been postulated to promote top-down cognitive-affective control; however, the existing clinical investigations show improvements in tinnitus outcomes without direct proof that LTP-like facilitation is the mechanism responsible. Therefore, frequency should not be considered generically inhibitory or excitatory. Its effect relies on the functional role and baseline state of the target, stimulation strength, number of pulses, train structure, intertrain interval, sequence and prior stimulation. Therefore, protocol optimization should define the coupling of stimulation frequency with the target cortical region and network-level goal.

4.3 Multisite paradigm for maintaining plasticity in tinnitus

Multisite stimulation is justified by the scattered character of persistent tinnitus. Single-site temporal stimulation may alter the sensory component, with relative preservation of attentional, salience, memory or emotional systems. Sequential multisite techniques thus aim to activate more than one component of the tinnitus network rather than to merely increase the number of activated locations. Clinical data is encouraging but not consistent. Kleinjung et al. (2008) observed early improvement following stimulation of the auditory cortex alone and sequential stimulation of the DLPFC and auditory cortex, but only the combination procedure led to persistent advantage at 3 months. Noh et al. (2019) showed that THI improved after combined auditory and prefrontal stimulation, with greater active-sham separation at intermediate follow-up, whereas Noh et al. (2020) reported sustained THI and VAS improvements after dual-site auditory-prefrontal stimulation when auditory-cortex-only stimulation was ineffective. These clinical findings are compatible with, but do not establish, coordinated regulation of sensory and regulatory network components. In contrast, Kreuzer et al. (2015) did not find the superiority of ACC → TPJ over DLPFC → TPJ stimulation, while Lehner et al. (2016) did not find substantial superiority of triple-site over single-site therapy. Thus, not necessarily more stimulation locations are associated with better effectiveness. The number of targets alone may be less essential than the biological significance of the chosen nodes, target sequence, laterality, dose distribution and interactions between subsequent stimulation blocks. Several trials also revealed delayed improvement followed by subsequent deterioration, suggesting that induction and maintenance might be separated. Booster or maintenance stimulation may have a function at some stage; however, maintenance programs should be considered only if a repeatable acute treatment impact has been proven and should include unambiguous reporting of cumulative pulses per target and treatment scheduling.

4.4 Importance of rTMS parameters in tinnitus

Therapeutic exposure is integrated regarding target, laterality, frequency, intensity, pulse number, train duration, intertrain interval, session number, target order and total treatment time. Most low-frequency procedures employed roughly 100%−110% of the resting motor threshold, 1,500–2,000 pulses each session, and 5–10 treatment sessions; however, other protocols used 3,000–4,000 pulses or stimulated numerous areas. Motor threshold calibration offers personalized stimulator output and safety but does not guaranty same cortical electric-field exposure at temporal or prefrontal targets because to variations in scalp-to-cortex distance, gyral shape, coil orientation, and tissue conductivity across regions. Higher pulse counts or longer treatment courses have sometimes been linked to greater or more persistent improvement, but the evidence available does not establish a clear dose–response threshold because pulse dose is confounded by frequency, target, train structure, sequence, session spacing, and participant phenotype (Müller et al., 2013; Schecklmann et al., 2015; Kreuzer et al., 2017). Therefore, for multisite procedures, the target-specific dosages should be reported separately rather than merged into a single cumulative result. Localization techniques varied from the worldwide 10–20 EEG system to MRI, fMRI, PET/MRI, tonotopic or pitch-matched navigation, and electric-field modeling. Image guiding increases anatomical repeatability, yet Sahlsten et al. (2019) found no obvious advantage of navigated compared to non-navigated therapy and some carefully focused investigations were negative. Therefore, structural accuracy does not guarantee functional target validity. Similarly, treatment-related alterations in regional metabolism, perfusion or hemodynamics should be regarded as mechanistic biomarkers rather than believed to be mediators of clinical improvement. For reproducibility, future studies should report the coil model and orientation, motor threshold method, target coordinates, localization procedure, estimated electric field if available, frequency, intensity, train and intertrain structure, pulses per target, target sequence, cumulative dose, number and spacing of sessions, adherence, adverse events and schedules of maintenance. In the absence of these characteristics, apparently comparable “1 Hz” or “dual-site” treatments might indicate quite distinct biological exposures.

4.5 State dependency and tinnitus

State dependency is still a relevant and independent factor, as chronic tinnitus is a dynamic network-level disease, not a fixed anomaly at one cortical site. Besides frequency, intensity and coil position, the physiological and therapeutic effects of a magnetic pulse may be influenced by the baseline activation, oscillatory state, cognitive state, emotional state and recent activity of the neuronal population being stimulated. Sack et al. (2024) stressed the relevance of brain state for target engagement. Silvanto and Pascual-Leone (2008) saw TMS effects as an interplay of external stimulation and the activation state of the targeted neurons. Therefore, a constant protocol might result in varied answers depending on the pathological condition of the auditory, attention, emotional and salience networks. This is especially true with consecutive stimulation. Prefrontal stimulation may change attention, emotional evaluation or top-down control before temporal stimulation is provided, hence affecting the network state exposed by the subsequent stimulation block. This gives a tested priming hypothesis, not an established mechanism of therapeutic benefit. Similarly, the psychological and affective state could alter the direction and amount of non-invasive brain-stimulation effects, suggesting that baseline distress and cognitive-emotional state might be significant treatment factors (Schutter et al., 2023). But state dependence cannot be the universal explanation for unfavorable research retroactively. Testing prospective. Future trials should pre-specify whether stimulation is delivered at rest or during an auditory, attentional, emotional, or cognitive task; document medication, sleep, recent sensory exposure, and baseline distress; and determine whether EEG, MEG, or functional connectivity measures modify treatment response. Studies of randomized sequences are also needed to evaluate if prefrontal priming actually affects the effects of subsequent stimulation of the auditory cortex.

4.6 Functional network state and biomarker-guided targeting

Functional target validity might be as essential as anatomical location. Tinnitus is associated with abnormality in the ascending auditory pathways, descending corticofugal control and widespread cortical networks. Cortical rTMS may change the cortical expression of abnormal auditory activity, but it does not directly remove all possible subcortical generators. Individual variations in network architecture may so restrict the efficacy of fixed anatomical targets. EEG and MEG research offer a possible framework for the identification of functionally important network states. Tinnitus has been associated with altered resting activity, reduced long-range alpha coupling, increased gamma coupling, associations between auditory-cortex gamma activity and subjective loudness, and distributed generators involving auditory, temporoparietal, insular, anterior cingulate and parahippocampal regions (Weisz et al., 2005; Schlee et al., 2009; van der Loo et al., 2009; Moazami-Goudarzi et al., 2010). Abnormal interactions between low- and high-frequency oscillations have also been documented in thalamocortical dysrhythmia models (Vanneste et al., 2018), whereas increases in alpha activity in the auditory cortex following rTMS have been related to reduced tinnitus loudness (Müller et al., 2013). These findings give a physiological basis for the prospective assessment of EEG- or MEG-informed target, frequency, timing and sequence selection. Candidate signals include alpha–gamma balance, localized temporal gamma activity, long-range connections, and post-stimulation alpha restoration are yet experimental and require independent confirmation. Future standardization may thus need to integrate anatomical neuronavigation, electric field estimate, individual auditory and psychological phenotype, and functional network state instead of using only a fixed scalp position.

4.7 Sham validity and methodological heterogeneity

Sham integrity is a crucial driver of projected efficacy as the main outcomes in tinnitus studies are largely patient-reported. Angled-coil sham may imitate sound without enough scalp feeling, sound-only sham may be easily recognizable, and specialized sham coils may lessen cortical stimulation but need good cutaneous sensory matching. In contrast, persistent physiologic stimulation during sham may diminish the apparent active-control difference. Hence, future sham systems should be independently evaluated. Allocation guesses of participants and operators should be registered together with blinding indices, coil orientation, sound attenuation, cutaneous stimulation, and estimated cerebral electric field. Outcome evaluation should be blinded when feasible, and treatment expectation evaluated. Active comparator studies are valuable for protocol improvement, but should not substitute physiologically inert and sensory believable sham comparisons when the primary concern is clinical effectiveness.

4.8 Clinical implications

Current data do not support rTMS as a regular first line or final therapy for chronic subjective tinnitus. Its usage should be regarded exploratory in selected patients, especially in registered trials or established clinical programs that examine therapy benefit, adverse effects and durability. Counseling, hearing assessment and rehabilitation, cognitive-behavioral techniques and appropriate sound-based treatment remain core elements of recognized care for tinnitus. Clinical interpretation should also reflect the specific outcome domain that was investigated. Auditory or temporoparietal stimulation is mostly focused on affecting the sensory-perceptual component, whereas DLPFC or ACC stimulation is focused on affecting attentional and emotional reinforcement; multisite techniques try to activate both. Thus, tinnitus loudness, distress, disability, functional effect, mood and quality of life should not be regarded as interchangeable therapeutic objectives. Reductions in THI or TQ do not necessarily reflect reduced tinnitus loudness, nor does temporary perceptual suppression necessarily suggest lasting improvement in tinnitus related impairment or daily functioning.

4.9 Conclusion

Current evidence does not support any single standardized rTMS approach for routine clinical treatment of persistent subjective tinnitus. Low-frequency auditory or temporoparietal stimulation has had mixed success, and high-frequency temporal stimulation has exhibited only transient perceptual suppression. Some research suggests that combined auditory–prefrontal and multisite protocols may be helpful, particularly to target both perceptual and tinnitus-related strain, although they have not consistently shown superiority to single-site stimulation. These conclusions represent the contrast between direct clinical data, physiologically feasible mechanistic explanations and emergent personalized techniques outlined in Supplementary Table S6. Response to therapy is expected to be moderated by cortical target, stimulation frequency and dosage, treatment sequence, tinnitus phenotype, hearing status, baseline network state, outcome selection and sham credibility. Where safety outcomes were reported, rTMS was generally feasible and well-tolerated; however, adverse-event reporting was inconsistent. Future trials should use standardized and comprehensively reported protocols, credible sensory-matched sham conditions, clearly defined outcome domains and responder criteria, systematic audiological and clinical phenotyping, formal risk-of-bias assessment, consistent safety reporting, and longer follow-up, while network- and biomarker-guided approaches may support individualized treatment but require prospective validation. Until persistent and substantial superiority over sham and standard tinnitus therapy is proven, rTMS should remain an exploratory neuromodulatory treatment for chronic subjective tinnitus.

4.10 Future directions

Future studies should shift from strict anatomical targeting to appropriately verified circuit- and phenotype-informed neuromodulation. Multi-center randomized studies should pre-register mechanistic theories, use independently verified active-sham systems, standardize THI, TQ or TFI measurement, specify clinically relevant responder criteria and prolong follow-up beyond the immediate post-treatment period. Factorial or head-to-head designs are needed to separate the effects of frequency, target, intensity, pulse dosage, session number, sequence, and maintenance approach. Stratification should take into account hearing status, tinnitus laterality and duration, baseline symptom load, somatic modulation, depression or anxiety, sleep, medication and electrophysiological phenotype. Prospectively, EEG-/MEG-guided stimulation, connectivity-informed targeting, electric-field modeling and closed-loop timing should be compared to traditional structural navigation. Digital twin techniques are an emerging computational tool for tailored target selection, but prospective clinical validation is needed to show therapeutic value. Similarly, metabolic, perfusion, hemodynamic or other neurophysiological variables should be prospectively correlated with pre-specified clinical outcomes and not be construed as established mechanisms of therapy. Two contextual reports not included in the 28-study primary synthesis are presented to demonstrate new directions. Low-frequency rTMS of the auditory cortex with cognitive-behavioral treatment has been reported by Richter et al. (2017). The data is restricted to one instance, but it suggests a possible adjuvant or priming function for rTMS. Zhang et al. (2026) employed a digital-twin brain architecture to find unique, state-specific stimulation targets, demonstrating the promise of network-informed personalized neuromodulation. Both papers are still hypothesis-generating and require prospective confirmation. Multimodal therapy is another major field of study. The combination of rTMS with cognitive-behavioral treatment may allow the exploitation of a window of plasticity in which cognitive or emotional learning can be reinforced, although controlled trials are needed. Similarly, other combinations such as low-level laser treatment, pharmacologic modulation, relaxation methods, hearing rehabilitation, and sound-based therapies require controlled studies to differentiate additive, synergistic, and non-specific effects. Standardized core outcome sets, reporting templates for protocols, open reporting of individual-level data and harmonized biomarker acquisition would increase comparability and enable future individual-participant-data analyses. Thus, the fundamental question is not only whether a single cortical coordinate may be activated, but which network state, target combination, sequence and dosage is optimal for a particular tinnitus phenotype.

5 Limitations

Limitations of this study include high clinical and methodological variability, small sample numbers in several studies, variation in tinnitus phenotype, inconsistency in definitions of responders, non-uniformity of results, inadequate reporting of sham, and short or varying follow-up. The qualitative synthesis is unable to quantify between-study heterogeneity or estimate a pooled treatment effect. The mechanistic discussion mixes clinical findings with computational, imaging, electrophysiological and preclinical evidence. Therefore, cellular and network mechanisms, such as DCN metaplasticity, inflammatory signaling, LTD-/LTP-like plasticity, state dependency, and network-level regulation, should be considered as biologically plausible explanatory frameworks, unless directly demonstrated in the respective clinical study. Contextual mechanistic and emerging reports were included solely to sustain interpretation of protocol development and future research directions and were maintained distinct from the primary evidence base of 28 studies. Audiological and tinnitus phenotypic characteristics, in particular hearing-loss configuration, tinnitus pitch, hyperacusis and use of hearing aids or sound-based therapy, were also reported inconsistently, restricting systematic investigation of these variables as possible modifiers of treatment response. Safety reporting was also uneven amongst trials, prohibiting meaningful assessment of the frequency and breadth of treatment-related adverse events.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fauot.2026.1955517/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Grant Searchfield, The University of Auckland, New Zealand

Reviewed by: Ivan C. Carmona-Tortolero, Virginia Commonwealth University, United States

Hyunjung An, Hallym University, Republic of Korea