Modeling and Removing Physiological Artifacts in Neuroimaging

This direction is merged into another direction: low signal-to-noise ratio (neural signal-to-noise ratio). Its last version stays readable and citable here.

Version 1 of 138 papers40 conceptsfrozen 2026-10-04

Functional neuroimaging aims to map the electrical and metabolic activity of the brain, but the brain does not exist in a vacuum. It is housed within a body that breathes, circulates blood, and fluctuates in arousal. These physiological processes create a persistent background of "noise" that often dwarfs the actual neural signals researchers wish to study.

The central challenge in this research direction is the development of computational frameworks to isolate, characterize, and remove these non-neural artifacts—or, in some cases, to recognize that the "noise" itself carries vital biological information.

The Anatomy of Physiological Noise

In modalities like functional Magnetic Resonance Imaging (fMRI) and functional Near-Infrared Spectroscopy (fNIRS), the signal measured is a proxy for neural activity, typically based on blood oxygenation. This makes the data inherently susceptible to any process that moves blood or changes its chemistry.

As noted by Lilla Zollei and William M. Wells III in Exploratory Identification of Cardiac Noise in fMRI Images, cardiac and respiratory cycles introduce rhythmic, highly autocorrelated signals that can obscure neural activation. These artifacts are not merely random fluctuations; they are structured interference. For instance, Zaineb Amor and colleagues have shown that breathing introduces dynamic magnetic field fluctuations that reduce the sensitivity of high-field MRI, making it difficult to detect subtle brain activity (Impact of $\Delta \textrm{B}_0$ field imperfections correction on BOLD sensitivity in 3D-SPARKLING fMRI data).

Beyond mechanical cycles, the autonomic nervous system also plays a role. Pınar S. Özbay demonstrated in Sympathetic activity contributes to the fMRI signal that "fight or flight" sympathetic bursts cause widespread vasoconstriction, dropping the fMRI signal across the entire brain. This suggests that much of the "global noise" in neuroimaging is actually a direct reflection of systemic physiology.

Computational Strategies for Signal Isolation

To "clean" this data, researchers have moved beyond simple linear filters toward sophisticated generative and decomposition models.

  1. Subspace Reconstruction: In the realm of EEG, where muscle movements and eye blinks create massive artifacts, Jose L Contreras-Vidal and Saurabh Prasad developed EEGReXferNet. This framework uses cross-subject transfer learning to reconstruct signal subspaces, effectively "learning" what clean neural activity looks like to filter out noise in real-time (EEGReXferNet: A Lightweight Gen-AI Framework for EEG Subspace Reconstruction).
  2. Hemodynamic Decomposition: For fNIRS, which is particularly sensitive to scalp blood flow, Marco A. Pinto-Orellana and colleagues introduced the Hemodynamic Decomposition Model (HDM). By breaking down complex signals into interpretable biological pulses, they significantly improved the accuracy of cognitive load detection (A hemodynamic decomposition model for detecting cognitive load using functional near-infrared spectroscopy).
  3. Deep Learning Decoders: Pablo Ortega and Aldo Faisal developed HemCNN, a deep learning architecture that decodes motor tasks from fNIRS without requiring the cumbersome baseline corrections usually needed to account for physiological jitter (HemCNN: Deep Learning enables decoding of fNIRS cortical signals in hand grip motor tasks).

When Noise Becomes the Signal

A major shift in this field is the realization that physiological "nuisance" signals often contain structured information about the brain's state.

Jingwei Li and B.T. Thomas Yeo argued in Topography and behavioral relevance of the global signal in the human brain that the "global signal"—long dismissed as a breathing artifact—actually reflects major brain networks and predicts cognitive abilities. Similarly, Shagun Ajmera showed that "infra-slow" fluctuations can predict the progression of Alzheimer’s disease with high accuracy (Infra-slow brain dynamics as a marker for cognitive function and decline).

This duality creates a methodological tension: researchers must decide whether to remove a signal to see local neural activity or to model it to understand the brain's broader physiological context. For example, Carlo Tessa and Gaetano Valenza used the interaction between fMRI and heart-rate variability to show that autonomic impairment is a core feature of early Parkinson's disease (Central modulation of parasympathetic outflow is impaired in de novo Parkinson's disease patients).

Clinical and Diagnostic Utility

The ultimate goal of refining these denoising techniques is to improve the reliability of neuroimaging as a clinical tool. Yang Wang and Andrew J. Saykin demonstrated that when physiological noise is properly managed, methods like Pulsed Arterial Spin Labeling (PASL) become highly reproducible biomarkers for long-term clinical trials (Regional reproducibility of pulsed arterial spin labeling perfusion imaging at 3T).

By isolating genuine neural oscillations from background biological variability, researchers like Yang Liu have been able to identify "gradual disturbance" patterns that track the progression of the Alzheimer’s spectrum (Gradual Disturbances of the Amplitude of Low-Frequency Fluctuations (ALFF) and Fractional ALFF in Alzheimer Spectrum). As these denoising models become more precise, the line between "noise" and "signal" continues to blur, revealing a more integrated picture of the human brain as a biological organ.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_modeling_and_removing_physiological_arti_a0d15955,
  title = {Modeling and Removing Physiological Artifacts in Neuroimaging},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/modeling-and-removing-physiological-artifacts-in-neuroimaging/rdr_a0d15955036d4ea1a97a9373ce18389b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_modeling_and_removing_physiological_arti_a0d15955_v1,
  title = {Modeling and Removing Physiological Artifacts in Neuroimaging},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/modeling-and-removing-physiological-artifacts-in-neuroimaging/rdr_a0d15955036d4ea1a97a9373ce18389b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}