Physiological Noise Correction in Functional Magnetic Resonance Imaging

Functional magnetic resonance imaging (fMRI) relies on the blood oxygen level dependent (BOLD) signal to infer neural activity, yet the recorded time series are invariably contaminated by physiological fluctuations. Cardiac pulsation, respiration-induced chest and head motion, low-frequency vasomotor oscillations and autonomic vascular innervation all introduce variance that can obscure or mimic neuronal signals. At high field strengths and in rapid acquisition paradigms, the relative contribution of physiological noise rises markedly, necessitating robust correction strategies. Broadly, these fall into model-based regressions—using recorded cardiac and respiratory traces to generate nuisance regressors (for example RETROICOR, respiratory volume per time and heart rate variability models)—and data-driven approaches such as independent component analysis (ICA) to identify and remove structured noise components. Recent advances in multiecho sequences, high-density sampling and time–frequency methods allow more precise characterisation of the spectral and temporal dynamics of physiological artefacts. Effective denoising enhances sensitivity to subtle BOLD changes, improves reproducibility of resting-state networks and refines clinical biomarkers. As fMRI extends into studies of cerebral autoregulation, glymphatic clearance and brain–body coupling, meticulous treatment of physiological confounds is indispensable for accurate interpretation and global comparability of results.

Research from Nature Portfolio

Innovative multimodal imaging has demonstrated that physiological noise arises from more than cardiac and respiratory cycles. A recent study using an MRI-compatible functional eye camera combined with video ophthalmoscopy revealed concurrent respiratory, cardiac and vasomotor pulsations at the eye surface and in the brain, enabling decomposition of these rhythms through Fourier analysis and guiding the design of tailored regressors. Separately, work on sympathetic vascular innervation has shown that transient subcortical arousal events and skin vascular tone fluctuations co-occur with widespread BOLD signal changes, identifying a hitherto underappreciated systemic driver. Incorporation of sympathetic activity measures into denoising pipelines has improved isolation of neuronally driven BOLD variance and prompted revisions to standard physiological models.

Physiological Noise Correction in Functional Magnetic Resonance Imaging publication trend

The graph below shows the total number of articles in physiological noise correction in functional magnetic resonance imaging across all publications each year (not limited to Nature Index journals).

Technical terms

BOLD signal: Blood oxygen level dependent contrast reflecting changes in deoxyhaemoglobin that serve as an indirect marker of neural activity.

RETROICOR: Retrospective image correction technique that uses concurrently recorded cardiac and respiratory waveforms to model and regress physiological noise from the fMRI time series.

Independent component analysis (ICA): Data-driven decomposition method that separates fMRI data into spatially independent components, enabling identification and removal of noise sources without external recordings.

Wavelet analysis: Time–frequency decomposition approach that characterises signal coherence across scales, allowing dynamic separation of physiological rhythms from neuronal fluctuations.

Nuisance regression: General linear model procedure in which measured physiological, motion or scanner parameters are included as regressors to remove structured noise from fMRI data.

References

1. Synchronous functional magnetic resonance eye imaging, video ophthalmoscopy, and eye surface imaging reveal the human brain and eye pulsation mechanisms. Scientific Reports (2024).
2. Characterizing systemic physiological effects on the blood oxygen level dependent signal of resting‐state fMRI in time‐frequency space using wavelets. Human Brain Mapping (2023).
3. Hand classification of fMRI ICA noise components. NeuroImage (2016).
4. Potential pitfalls when denoising resting state fMRI data using nuisance regression. NeuroImage (2016).
5. Sympathetic activity contributes to the fMRI signal. Communications Biology (2019).