Partitioning peripheral from central signal in task-related EMG-fMRI of movement disorders

Question: How has task-related EMG-fMRI co-registration been used to distinguish neural drive from motion or hemodynamic artifact in dystonia and other movement disorders?

Answer

Across the retrieved literature, task-related EMG has functioned less as a classifier that certifies a BOLD effect as "neural" and more as a variance-partitioning instrument with two opposite uses. In the healthy motor system it is used constructively: adding EMG (and accelerometry) to a task model explains BOLD variability in primary sensorimotor cortex and, beyond accelerometry, in contralateral thalamus and secondary somatosensory cortex, which is read as increased specificity and as sensitivity to unpredicted movement [10]. In dystonia it is used deflationarily: in cervical dystonia the surface EMG time course from hand/arm and cervical muscles closely matched the contralateral BOLD time courses that carry the reported abnormality, including escalation across tapping blocks and sustained activity into rest [1]. Taken together, the papers support a narrower conclusion than the question presupposes: EMG-fMRI co-registration in this literature has demonstrated that peripheral muscle activity is a plausible source of dystonia-relevant BOLD signatures, but no retrieved study actually performs simultaneous, within-scan EMG-fMRI in dystonia patients and shows which effects survive EMG regression. Motion and hemodynamic correction, meanwhile, proceed on entirely separate toolchains, so the separation of neural drive from artifact remains asserted piecewise rather than jointly.

EMG regressors buy specificity mainly where movement is spontaneous or unpredicted

In a block-design ankle dorsi-/plantarflexion and gait-imagery paradigm, adding either accelerometry or tibialis anterior EMG on top of the task regressor robustly explained BOLD variability in primary sensorimotor cortices, and adding EMG on top of accelerometry additionally explained variability in the contralateral thalamus and secondary somatosensory cortex [10]. That incremental EMG effect was, however, mainly driven by spontaneous signal fluctuations at rest and during imagery rather than by the instructed contraction, and inclusion of a temporal derivative improved detection in primary sensorimotor cortex while potentially harming detection in subcortical regions [10]. The practical claim the authors draw is that EMG-informed modelling increases specificity and permits differentiation of spontaneous or unpredicted motor behaviour from task-driven activity [10].

In cervical dystonia, the BOLD signature used to define the pathophysiology has a peripheral twin

In cervical dystonia, BOLD signal after finger tapping was abnormally sustained in left putamen and left cerebellum, elevated during tapping, and escalated as the task repeated, with patient–control differences confined to the left hemisphere [1]. Surface EMG recorded during the same left-hand tapping protocol across hand/arm and cervical muscles bore a strong resemblance to the contralateral BOLD time courses at both group and individual level, reproducing both the escalation during tapping and the sustained activity during post-tapping rest blocks, with the largest and most consistent effects in the right sternocleidomastoid [1]. Because the EMG match extends to the post-movement rest period that constitutes the reported abnormality, the study cannot exclude that sustained putaminal and cerebellar BOLD in cervical dystonia partly reflects sustained cervical muscle activity rather than a central after-effect [1].

Dystonia connectivity studies still anchor muscle representations with task localisers, not EMG

In laryngeal and focal hand dystonia, individually generated hand and mouth/larynx regions were derived from task-based MRI (finger tapping and phonation respectively) and used to interrogate resting-state connectivity [2]. Both subtypes showed a shared interaction effect driven by higher connectivity between the somato-cognitive action network and the task-derived mouth/larynx region, with concomitantly lower connectivity to the cingulo-opercular network [2]. The task localiser here defines the effector representation but provides no measure of whether patients and controls generated equivalent muscle output, so the co-registration is behavioural-by-proxy rather than electromyographic.

When the symptom is the movement itself, kinematics have served as the covariate

In writer's cramp, writing was performed outside the scanner with a kinematic device and repeated during fMRI; group differences in functional network activity during writing were absent, but reduced left sensorimotor and superior parietal network activity correlated with greater writing dysfluency indexed by peak acceleration, a pattern distinct from healthy volunteers [3]. Task-based fMRI acquired during a writing task likewise showed that 10 Hz rTMS to primary somatosensory cortex, but not premotor cortex, improved writing dysfluency and weakened cortico-basal ganglia, cortico-cerebellar and intra-cerebellar connectivity while strengthening striatal connectivity [11]. In dyskinesia, kinematics-defined movement presence separated four behavioural states and significantly affected the predictive performance of oscillatory biomarkers, with a movement-aware classifier improving dyskinesia detection [5]. In each case an objective peripheral measure of the abnormal movement, not EMG, does the work of distinguishing behaviour from neural state.

Motion and physiological noise are corrected on a separate toolchain whose residuals are not EMG-shaped

Inertial navigation exploiting the scanner's static magnetic and gravitational fields estimated 3-DOF head orientation at 2 kHz without integration, and with phase-correlation for translation achieved 0.6° and 0.4-pixel accuracy, improving image quality for 2D EPI fMRI in both retrospective and prospective correction during large volunteer motions [7]. For physiological noise, RETROICOR, heart rate and respiratory volume explained significant variance throughout the brain with peaks around areas of strong cardiac pulsation; ICA-AROMA and aCompCor accounted for a substantial proportion of the variance explained by these peripheral methods but the peripheral methods retained unique explanatory power, and ICA-AROMA was unreliable at consistently identifying and removing physiological noise across recordings within participants [12]. In spinal cord fMRI, PCA-derived nuisance regressors from a noise region fitted noise but did not clearly benefit group-level maps, and collinearity of these regressors with the task was flagged as a concern for implementations where task-correlated noise is anticipated; such denoising is recommended in place of physiological recordings only when the latter are unavailable [13]. Task-correlated muscle activity is precisely that anticipated case, and none of these methods takes EMG as input.

Muscle is itself a BOLD source, which undermines rest-versus-task designs in dystonia

Regional homogeneity applied to resting-state fMRI of calf muscle detected significant differences between active and inactive muscles during isometric plantar and dorsal flexion without any rest–activity contrast, motivated explicitly by the unsuitability of conventional rest-versus-activity comparisons for patients with sustained muscle contractions such as spasticity and dystonia [6]. This establishes that contracting muscle generates locally synchronised BOLD signal measurable in a single series, and that in disorders where a clean muscular baseline does not exist, the reference state assumed by subtractive designs is itself unavailable [6].

Conflicts and gaps

The same EMG evidence is read in opposite directions. One paper treats EMG-explained variance as noise whose removal increases specificity, while noting that the incremental variance comes from spontaneous movement [10]; the other treats EMG–BOLD resemblance as grounds for doubting a patient–control BOLD difference [1]. Which reading applies to a given effect is unresolved.

No retrieved study performed simultaneous EMG-fMRI in dystonia patients. The cervical dystonia EMG was acquired from a subset of patients and controls at a follow-up session rather than during scanning [1], so the linkage is across sessions and cannot serve as a within-run nuisance regressor. The only EMG-informed GLM identified used healthy participants and a lower-limb task [10], leaving cervical, laryngeal and facial muscles — where susceptibility artefacts and head motion couple tightly to muscle activity — untested.

Whether peripheral physiological recordings or data-driven regressors are preferable remains contested: peripheral recordings were reported to retain unique explanatory power [12], while data-driven PCA regressors failed to improve group maps and risk task collinearity [13]. No retrieved paper validates any correction against a ground truth in which muscle activity is known to be absent.

What would change the answer

Simultaneous, multi-muscle surface EMG during fMRI in cervical or laryngeal dystonia, entering the EMG (and its temporal derivative) as a nuisance regressor alongside task and motion parameters, and reporting explicitly whether the sustained putaminal and cerebellar effects survive — the design the existing follow-up-session data only approximate [1] [10].

Botulinum toxin as a peripheral de-efferentation control is the cleanest available manipulation: imaging the same patients at peak effect and at end of cycle, as done for 7T digit mapping in focal hand dystonia [4], would show whether a BOLD abnormality tracks paralysis or persists unchanged with equivalent intended task.

Motor-unit decomposition offers a genuine neural-drive signal rather than a compound muscle signal: biophysical-model-informed source separation estimates neural drive and motor-neuron properties from surface EMG [9], and motor-unit beta bursts in forearm extensors are time-locked to cortical beta bursts [8]. Decomposed drive, as opposed to raw EMG amplitude, is the regressor that could in principle separate descending command from peripheral re-afference.

Finally, combining EMG with high-rate inertial head tracking [7] would break the collinearity between neck-muscle activation and head displacement that makes cervical dystonia the hardest case, and a head-to-head comparison of RETROICOR-plus-EMG against aCompCor and ICA-AROMA under conditions of deliberate task-correlated muscle activity [12] [13] would test whether EMG adds anything beyond existing denoising.

References

[1] Kuster, J. K., Levenstein, J. M., Waugh, J. L., Multhaupt-Buell, T. J., Lee, M. J., Kim, B. W., Pagnacco, G., Sapienza, S., Bonato, P., Makhlouf, M. L., Sudarsky, L. R., Breiter, H. C., Sharma, N., Blood, A. J.. "Sustained activation in basal ganglia and cerebellum after repetitive movement in cervical dystonia". bioRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.1101/2023.03.19.533030

[2] Wang, Y., Huynh, B., Ren, J., Chen, M., Zhang, W., Hu, D., Li, S., Liu, H., Kimberley, T. J.. "Somato-cognitive action network in laryngeal and focal hand dystonia sensorimotor dysfunction". medRxiv. 2025-03-07T00:00:00.000Z. URL: https://doi.org/10.1101/2025.02.21.25322612

[3] Bukhari-Parlakturk, N., Mulcahey, P. J., Fei, M., Lutz, M. W., Voyvodic, J. T., Davis, S. W., Michael, A. M.. "Increased sensorimotor and superior parietal activation correlate with reduced writing dysfluency in writer's cramp dystonia". medRxiv. 2025-03-21T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.20.25324331

[4] Asghar, M., Francis, S., Sanchez-Panchuelo, R., Glover, P., Pakenham, D., O'Neill, G. C., Schluppeck, D., Humberstone, M.. "Assessing the digit organisation of focal hand dystonia using 7T functional MRI". bioRxiv. 2025-09-17T00:00:00.000Z. URL: https://doi.org/10.1101/2025.09.11.675579

[5] Habets, J. G., Merk, T., Mathiopoulou, V., Lofredi, R., Kaplan, J., Binns, T. S., Busch, J. L., Koehler, R. M., Memarpouri, A., Faust, K., Krause, P., Schneider, G.-H., Neumann, W.-J., Tovote, P., Kuehn, A. A.. "Movement execution defines a distinct neural state in dyskinesia and enhances decoding". bioRxiv. 2025-08-25T00:00:00.000Z. URL: https://doi.org/10.1101/2025.06.09.658641

[6] Hubena, T., Krupicka, R., Jech, R.. "Regional Homogeneity in Muscle Resting-State fMRI as a Novel Marker of Muscle Activation". medRxiv. 2025-09-12T00:00:00.000Z. URL: https://doi.org/10.1101/2025.09.11.25335521

[7] Musa Tunc Arslan, Fatih Calakli, Joshua Auger, Hongli Fan, Alan J Macy, Simon K Warfield. "MR-Compass: Inertial Navigation-Driven Motion Correction for Brain MRI". arxiv. 2026-03-03T00:00:00.000Z. URL: https://arxiv.org/abs/2603.01584

[8] Graef, C., Valdunciel, A. P., Farina, D., Vaidyanathan, R., Tai, Y. F., Haar, S.. "Propagation of Beta bursts from the motor cortex to the motor units of multiple upper-limb muscles". bioRxiv. 2025-05-27T00:00:00.000Z. URL: https://doi.org/10.1101/2025.05.22.655541

[9] D. Halatsis, P. Mamidanna, J. Pereira, D. Farina. "A Biophysical-Model-Informed Source Separation Framework For EMG Decomposition". arxiv. 2025-10-22T00:00:00.000Z. URL: https://arxiv.org/abs/2510.17822

[10] Jasenska, M., Hok, P., Kojan, M., Burkot, O., Kolarova, B., Holobar, A., Hlustik, P.. "Precise modeling of task-related sensorimotor activation based on simultaneous surface electromyography". medRxiv. 2026-02-03T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.01.26345279

[11] Bukhari-Parlakturk, N., Mulcahey, P. J., Lutz, M., Ghazi, R., Huang, Z., Dannhauer, M., Termsarasab, P., Scott, B., Simsek, Z. B., Groves, S. N., Lipp, M., Fei, M., Tran, T. K., Wood, E., Beynel, L., Petty, C., Voyvodic, J. T., Appelbaum, L. G., Al-Khalidi, H. R., Davis, S. W., Michael, A. M., Peterchev, A. V., Calakos, N.. "Motor network reorganization associated with rTMS-induced writing improvement in writer's cramp dystonia". medRxiv. 2025-01-18T00:00:00.000Z. URL: https://doi.org/10.1101/2024.09.30.24314652

[12] Krentz, M., Tutunji, R., Kogias, N., Mahadevan, H. M., Reppmann, Z., Krause, F., Hermans, E. J.. "Physiological Noise Correction in Brainstem Imaging: An Empirical Evaluation of fMRI Data-Driven and Peripheral Physiological Recording-Driven Methods". bioRxiv. 2025-09-23T00:00:00.000Z. URL: https://doi.org/10.1101/2023.02.22.529506

[13] Hemmerling, K. J., Vigotsky, A. D., Glanville, C., Barry, R. L., Bright, M. G.. "Data-driven denoising in spinal cord fMRI with principal component analysis". bioRxiv. 2026-01-28T00:00:00.000Z. URL: https://doi.org/10.1101/2025.01.23.634596

How this survey was made

This survey answers one question: “How has task-related EMG-fMRI co-registration been used to distinguish neural drive from motion or hemodynamic artifact in dystonia and other movement disorders?”

We keep our own collection of recent preprints and refresh it every day from bioRxiv, medRxiv, arXiv, PsyArXiv, SocArXiv and ChemRxiv. For each survey, we search that collection for recent work and keep the strongest matches: papers that use the same terms as the question, and papers on the same topic in different words.

The text was written from the titles and abstracts of those papers; the full texts were not read. Statements that draw on a particular paper are marked with a number in brackets. The reference list just above is the 13 papers whose work this text draws on, chosen from the 114 papers those searches returned; the rest of the shortlist is not printed. It is not a complete review of everything published on the question.

Treat this as a starting point and a reading list, not a peer-reviewed review. The searches behind this survey ran on October 21, 2026 and covered work from the previous two years.