The technique in question is repetitive transcranial magnetic stimulation, or rTMS, a procedure in which rapidly pulsed magnetic fields are delivered through a coil held against the scalp. The magnetic pulses pass painlessly through the skull and induce small electrical currents in the underlying cortical tissue, nudging neurons toward greater or lesser activity depending on the stimulation parameters. Clinicians have long used rTMS targeting the primary motor cortex, the strip of brain tissue at the top of the head that commands voluntary movement, to treat a range of motor symptoms. What has remained frustratingly unclear is precisely how the stimulation remodels the brain’s locomotor circuitry when it is applied to freezing of gait, a symptom that does not respond to medication in the same way tremor or rigidity often do.

To answer that question, the research team assembled a carefully structured cohort. They enrolled 37 patients with Parkinson’s disease who experienced freezing of gait, 66 patients with Parkinson’s disease who did not, and 32 healthy controls. Every participant underwent a comprehensive baseline assessment, designated T0, combining transcranial magnetic stimulation paradigms with multimodal magnetic resonance imaging. This dual approach allowed the investigators to capture both the electrophysiological behavior of the cortex and the structural and functional architecture of the brain in the same individuals, on the same day, creating a detailed neurophysiological fingerprint of what distinguishes patients who freeze from those who do not.

The baseline findings were striking. Compared with both the non-freezing patients and the healthy controls, the freezing group showed decreased short-interval intracortical inhibition, a measure of how effectively inhibitory circuits within the motor cortex damp down excessive neural activity. In essence, the motor cortex of patients who freeze is running too hot, with its internal braking systems weakened. The same group also displayed excessive cortico-cortical coupling, meaning their motor cortex was functionally entangled with other cortical regions in an abnormally strong way, and reduced fractional anisotropy across extensive white matter bundles, a diffusion MRI marker indicating that the insulated fiber tracts connecting distant brain regions had lost structural integrity. Together, these three abnormalities paint a picture of a locomotor control system that is disinhibited, mis-wired, and structurally compromised.

With that baseline established, the researchers turned to treatment. The 37 patients with freezing of gait received ten add-on rTMS sessions targeting the lower limb region of the primary motor cortex, the portion of the motor map that controls the legs and feet. The stimulation protocol used a frequency of 10 hertz, meaning ten pulses per second, delivered at 90 percent of each patient’s resting motor threshold, the stimulation intensity calibrated individually to the minimum needed to evoke a measurable muscle response. Delivering high-frequency stimulation at this sub-maximal intensity is a standard approach for exciting the targeted cortex while keeping the procedure safe and tolerable. Because the sessions were added on top of existing care, the study could evaluate whether stimulation provided benefit beyond the patients’ usual treatment.

The results, measured one day after the final session and again at a one-month follow-up, showed meaningful improvements in objective gait indicators at both time points compared with baseline. This persistence is important: a therapy that only works for a few hours after the last session would be of limited clinical value, but the fact that gait measures remained improved a month later suggests that the stimulation produced changes that outlasted the treatment course itself. For a symptom that has historically been considered one of the most treatment-refractory features of Parkinson’s disease, the durability of the effect is a headline finding in its own right.

Just as importantly, the post-intervention assessments revealed what was happening inside the brain as patients improved. The team documented a constellation of neuromodulatory changes that partially reversed the abnormalities identified at baseline. Short-interval intracortical inhibition increased, indicating that the weakened inhibitory circuits within the motor cortex had been strengthened by the stimulation. Functional connectivity between the primary motor cortex and the occipital lobe, the visual processing region at the back of the brain, decreased, suggesting that the excessive cortico-cortical coupling seen before treatment had been dialed back. Perfusion, the blood supply reflecting neural activity, was reduced in the left precentral gyrus, the cortical region housing the motor representation of the body. And fractional anisotropy rose in the superior longitudinal fasciculus, a major white matter highway that links frontal, parietal, and temporal regions and plays a role in integrating motor planning with sensory feedback.

These mechanistic findings matter because freezing of gait has long resisted a simple explanation. It does not map cleanly onto the loss of dopamine that defines Parkinson’s disease, and it often persists even when other motor symptoms are well controlled by medication. The new data support a model in which freezing emerges from a network-level failure: a motor cortex that lacks adequate local inhibition, communicates abnormally with distant cortical areas including visual regions, and sits atop white matter pathways that can no longer relay signals efficiently. By strengthening cortical inhibition, normalizing the motor-occipital loop, reducing excessive perfusion, and remodeling the microstructure of key fiber tracts, rTMS appears to push this dysfunctional network back toward a healthier operating state.

The study also demonstrates the power of combining electrophysiology with multimodal imaging in a single experimental design. Transcranial magnetic stimulation paradigms provide a direct, quantitative readout of cortical circuit behavior in living humans, while diffusion and functional MRI reveal the structural scaffolding and functional coordination of the wider network. Measuring both before and after intervention allowed the researchers to link symptom change to specific, measurable neural mechanisms rather than relying on subjective reports alone. This kind of mechanistic transparency is increasingly seen as essential for moving brain stimulation from an empirical trial-and-error therapy toward a rationally targeted intervention.

There are, of course, limits to what a single study can establish. The intervention was evaluated in a defined cohort, and the authors note that the modulation of the identified abnormalities was partial rather than complete, meaning that stimulation improved but did not fully normalize the network disturbances underlying freezing. Larger and longer studies will be needed to determine optimal stimulation parameters, identify which patients are most likely to benefit, and confirm how long the effects can be sustained with maintenance treatment. Nevertheless, the work represents a significant step forward. It validates the primary motor cortex as a stimulation target for freezing of gait, identifies the cortical, connectivity, perfusion, and white matter changes that accompany clinical improvement, and offers a mechanistic framework that could guide the next generation of neuromodulation trials for one of Parkinson’s disease’s most feared symptoms.

Subject of Research: Neuromodulatory effects of repetitive transcranial magnetic stimulation on freezing of gait in Parkinson's disease

Article Title: Neuromodulatory mechanism of repetitive transcranial magnetic stimulation for freezing of gait in Parkinson’s disease

Article References: Gan, C., Sun, H., Cao, X., Si, Q., Wan, C., Ye, S., Shan, A., Gao, M., Shi, J., Wang, X., Ge, Y., Yuan, Y., Kong, Y., & Zhang, K. (2026). Neuromodulatory mechanism of repetitive transcranial magnetic stimulation for freezing of gait in Parkinson’s disease. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02800-7

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02800-7

Keywords: Parkinson's disease, freezing of gait, transcranial magnetic stimulation, primary motor cortex, cortical excitability, multimodal MRI, white matter integrity, brain stimulation, neuromodulation, movement disorders, gait analysis, neurology

The framework, called Region-Adaptive Multimodal Learning, or RAML, was developed by Naga Maha Lakshmi K. and Durgesh Nandan of SR University in Warangal together with Sushma Parihar of Symbiosis Institute of Technology in Pune. It builds on the well-known Attention U-Net encoder–decoder, but inserts two architectural stages that the authors say distinguish it from prior region-aware methods. First, an auxiliary set of 1×1 convolutional heads produces coarse probability maps for the three clinically relevant subregions: whole tumor, tumor core, and enhancing tumor. Second, the decoder feature map is explicitly split, gated, and re-fused per subregion before the final prediction head, so that subregion conditioning becomes part of the network’s forward pass rather than a training-time afterthought. A soft nesting penalty enforces the anatomical hierarchy that enhancing tumor must sit inside tumor core, which must sit inside whole tumor.

The second and arguably more novel stage is morphology-aware feature modulation. From the network’s own soft, sigmoid-based subregion masks, RAML computes three geometric descriptors per axial slice: region size, compactness, and boundary complexity. Crucially, these descriptors are computed on the soft masks rather than on thresholded, binarized versions, which keeps the entire descriptor pipeline differentiable, so gradients can flow from the modulation weights back through the descriptors into the auxiliary heads. The descriptor vector is passed through a small multilayer perceptron that outputs channel-wise scaling weights, which then modulate the refined feature map. Because early-training masks are unreliable, the modulation strength ramps linearly from zero to one over the first ten epochs, preventing noisy masks from injecting harmful gradients. The gating also operates between the refined features and a learned 1×1 bypass projection rather than between a feature map and an identical copy of itself, a design detail that keeps the module’s contribution verifiably non-trivial.

Training followed a 2.5D strategy: axial slices containing tumor, plus a small random sample of background-only slices, resized to 224 by 224 pixels, with the four co-registered modalities stacked as input channels. The model was trained end-to-end on an NVIDIA A100 GPU using the AdamW optimizer, a cosine annealing learning-rate schedule, mixed-precision arithmetic, and a composite loss combining Dice, cross-entropy, auxiliary subregion supervision, and the nesting penalty. All hyperparameters were selected by grid search on a validation set, and the test split of 55 patients was held out entirely from training, checkpoint selection, and tuning.

The reported results are notable less for what RAML wins than for how honestly the authors present what it does not. Across three independent random seeds on the held-out test set, RAML achieved a mean Dice of 0.9007 for whole tumor, 0.8321 for tumor core, and 0.7572 for enhancing tumor. That is below U-Net, TransUNet, and Swin-UNETR, which outperformed it by 1.4 to 5.6 percentage points depending on region and baseline. The authors report that this gap is statistically robust, with Holm-adjusted p-values below 0.005 across all three 2D baselines and all regions, based on paired Wilcoxon signed-rank tests over pooled per-patient scores from 155 to 165 observations per comparison. They also disclose that an earlier version of the work contained higher, unverified numbers caused by a data-pipeline problem that has since been corrected, an unusual and commendable act of transparency in a field often criticized for inflated benchmarks.

Where RAML does compete is efficiency. The model contains just 1.64 million parameters, roughly 4.7 to 13.2 times fewer than the evaluated baselines, which range from 7.76 million for U-Net to 21.68 million for TransUNet. It retains about 96.0 to 96.3 percent of the baselines’ average Dice performance while recording the lowest computational load of any model compared, at 3.72 GFLOPs per inference against 9.32 to 12.21 GFLOPs for the 2D baselines, and the second-lowest peak inference memory at 205.1 megabytes. Interestingly, that frugality does not translate into the fastest wall-clock speed: U-Net ran roughly twice as fast at 174.9 inferences per second versus RAML’s 95.2, likely because the region-adaptive and morphology modules perform sequential operations on small tensors that parallelize poorly on GPU hardware compared with the baselines’ larger, more uniform convolutional blocks.

The comparison against a true 3D baseline is the most encouraging part of the evaluation. Against a standard 3D U-Net with about three times RAML’s parameter count, the two models were statistically indistinguishable on whole tumor Dice, and the 3D model’s advantage on tumor core and enhancing tumor was modest. In five of nine region-metric combinations the differences were not significant. This suggests that the 2.5D slice-based design, often assumed to sacrifice meaningful volumetric context, costs RAML surprisingly little accuracy, an empirical point the authors support with an inter-slice consistency analysis showing continuity gaps of only 0.009 to 0.015 Dice relative to ground truth between adjacent predicted slices.

The ablation study adds nuance rather than a simple success story. Comparing a backbone-only variant against versions with region-adaptive refinement, morphology modulation, both, or both without the nesting loss, the authors found that the full configuration did not uniformly beat every simpler variant; the complete RAML scored slightly lower on enhancing tumor Dice than two of the component subsets. The absolute differences were small and, without formal significance testing, cannot be read as proof of one configuration’s superiority. What the ablation did establish clearly is that the architectural fusion itself, rather than the nesting loss alone, drives the near-elimination of anatomical hierarchy violations: the backbone-only variant violated the ET-subset-TC-subset-WT constraint on 39.0 percent of relevant voxels, while all architecturally complete variants dropped to between 0.0007 and 0.011 percent.

Perhaps the most methodologically instructive finding is a negative one. The team attempted to validate generalization on BraTS2018 as an independent external cohort, but cross-referencing the official BraTS inter-year patient mapping revealed that all 285 patients in their copy of the 2018 release already appear in the 2020 training cohort of 368 patients, leaving zero eligible non-overlapping subjects. Any accuracy figures computed on BraTS2018 after training on BraTS2020 would therefore constitute train-test leakage rather than genuine generalization. By reporting this openly, the authors flag a pitfall that likely affects other published cross-year evaluations of BraTS data, and they identify evaluation on a truly independent cohort with different scanners and acquisition protocols as a priority for future work.

The authors are careful about what RAML is and is not. It is not a state-of-the-art accuracy result, and it is certainly not a clinically validated or deployment-ready system; establishing clinical readiness would require prospective, multi-site validation and regulatory-grade evaluation. What the study does deliver is a verified, multi-seed, patient-wise demonstration that conditioning feature representation on tumor subregion identity and explicit geometric descriptors can approach the accuracy of models several times its size while cutting parameters by an order of magnitude and FLOPs by a factor of two to three. For clinical settings with limited computational resources, and for the broader effort to make medical AI models small enough to run at the point of care, that trade-off, a few accuracy points surrendered for a dramatic reduction in model complexity, may prove to be the most consequential finding of all.

Subject of Research: Region-adaptive multimodal deep learning for multimodal MRI brain tumor segmentation

Article Title: Region-adaptive multimodal learning with morphology-guided feature modulation for brain tumor segmentation

Article References: K., N. M. L., Nandan, D., & Parihar, S. (2026). Region-adaptive multimodal learning with morphology-guided feature modulation for brain tumor segmentation. Discover Artificial Intelligence, 6(1), Article 1350. https://doi.org/10.1007/s44163-026-02348-z

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02348-z

Keywords: brain tumor segmentation, multimodal MRI, deep learning, Attention U-Net, morphology-aware modulation, BraTS, Dice coefficient, medical image analysis, glioma, model efficiency, transformer baselines, 2.5D segmentation