Brain Biomarkers Predict TMS Success for Tinnitus

A specific brain region’s size may predict whether a person with tinnitus will benefit from repetitive transcranial magnetic stimulation (rTMS). In a new study, researchers found that patients who responded well to rTMS treatment had a significantly larger volume of gray matter in the right pars triangularis of the inferior frontal gyrus—a part of the brain involved in cognitive control and attention—before treatment even began. This structural feature emerged as the top predictor in a machine learning model designed to forecast rTMS outcomes.

A Search for Predictors in a Variable Treatment

rTMS is a non-invasive brain stimulation technique sometimes used to treat chronic tinnitus. It aims to modulate abnormal neural activity in brain networks linked to sound perception and attention. However, its effectiveness varies greatly from person to person; some experience significant relief while others notice little change. This variability, coupled with the cost and time commitment of treatment, has driven the search for biomarkers—measurable indicators that can predict who is most likely to benefit.

Led by researchers Zhongling Ding, Bo Peng, and Mengfang Gong, the new study hypothesized that pre-existing structural differences in the brain might hold the key. Given that tinnitus is associated with measurable changes in brain structure and that rTMS works by inducing neuroplastic changes, the team reasoned that certain brain architectures might be more receptive to the treatment’s effects.

How the Study Identified a Brain Biomarker

The researchers prospectively enrolled 64 patients with subjective tinnitus and 18 healthy controls. All tinnitus patients underwent a two-week course of rTMS treatment. To measure outcomes, they used standard clinical tools: the Tinnitus Handicap Inventory (THI) and a Visual Analogue Scale (VAS) for tinnitus loudness and distress. Patients showing significant improvement on these scales were classified as responders.

Before treatment began, each participant received a high-resolution structural MRI (sMRI) scan. The researchers extracted 242 distinct measurements of brain morphology from these scans, covering cortical thickness, surface area, and gray matter volume across the entire brain.

Using univariate statistics and then machine learning, they compared the brain scans of responders and non-responders to find features that differed between the groups. The most predictive features were fed into an “ExtraTreesGini_BAG_L1” machine learning model, which was rigorously tested using 5-fold cross-validation to ensure its reliability. The model’s decisions were then interpreted using SHAP analysis, a method that identifies which features contributed most to the predictions.

The Right Frontal Brain Region Emerges as Top Predictor

Thirty-six of the 64 patients (56.25%) were classified as responders to rTMS. The analysis revealed ten regional brain features that distinguished responders from non-responders, spanning networks involved in prefrontal, limbic, sensorimotor, and parietal functions.

The machine learning model built on these features performed robustly, with an area under the curve (AUC) of 0.85 and an accuracy of 0.77. Most importantly, SHAP analysis pinpointed a single feature as the most influential for a positive prediction: the gray matter volume (GMV) of the right pars triangularis of the inferior frontal gyrus (IFGtriang-R). A larger volume in this region made the model more likely to classify a patient as a future responder.

Further comparison showed that this finding was specific. The IFGtriang-R GMV was significantly larger in responders (0.90 ± 0.08) than in both healthy controls (0.86 ± 0.06) and non-responders (0.86 ± 0.07). This suggests responders possess a unique structural signature, not merely a “healthier” or more “normal” brain, but a specific enlargement in an area critical for cognitive control and attention switching.

Interestingly, the size of the IFGtriang-R did not linearly correlate with the degree of improvement (ΔVAS or ΔTHI), nor did any structural feature strongly correlate with baseline tinnitus severity. This points toward a threshold effect: having a certain amount of volume in this region may create a neuroplastic reserve that makes the brain responsive to rTMS, but more volume beyond that threshold doesn’t necessarily equate to better outcomes.

Toward Precision Neuromodulation for Tinnitus

The practical implication of this research is significant. A pre-treatment sMRI scan focusing on key regions like the IFGtriang-R could become a tool for patient stratification. Clinicians could potentially identify individuals with a high probability of responding to rTMS, thereby increasing the treatment’s overall success rate and saving other patients from an ineffective and costly procedure.

This work aligns with a broader movement toward personalized medicine in auditory health. Just as research explores how combined brain stimulation and therapy can improve outcomes, or how AI is applied in hearing health, using biomarkers to guide treatment selection represents a logical step forward. It moves away from a one-size-fits-all approach and toward what the authors call “precision neuromodulation.”

The study also deepens our understanding of tinnitus itself. The involvement of the right inferior frontal gyrus underscores the role of non-auditory brain networks—particularly those governing attention and executive control—in maintaining the tinnitus percept. Successful treatment may depend on an individual’s inherent capacity in these networks to be modulated.

For patients and clinicians, this research, detailed in the paper “Structural MRI biomarkers predictive of rTMS efficacy in subjective tinnitus” (Front. Neurol., 2026), offers a glimpse into a future where treatment pathways are informed by objective brain measures. While more validation in larger groups is needed, it provides a clear scientific direction for improving the application of neuromodulation therapies for tinnitus.