What's Doing the Work? Isolating the Active Ingredients in ILF Neurofeedback

Fred Shaffer
4 days ago
8 min read

Updated: 3 days ago

This post explains and evaluates de Matos and colleagues' (2026) disassembly study of infra-low neurofeedback, developed by the Othmers in 2006.

Understanding the Science of ILF Neurofeedback

Neurofeedback belongs to the larger family of biofeedback because it transforms a physiological signal into a real-time experience that a person can learn from. In this study, the signal was electroencephalography (EEG), and the learning context was deliberately subtle: an implicit neurofeedback format in which participants were simply instructed to watch an animation without trying to control it. The scientific bet was that the brain can still respond to feedback contingencies even when the task is passive, which makes the method easier to tolerate and closer to how many clinical sessions are actually delivered.

"The key scientific question is not whether neurofeedback works in general, but what exactly is doing the work."

The authors focused on a specific clinical approach called infra-low-frequency neurofeedback (ILF-NFB), which typically blends two feedback streams at once. One stream summarizes activity in the familiar classic frequency bands, while the other represents extremely slow fluctuations, the infra-low frequency component. Clinically, these are often delivered together, yet the field has not had a clean mechanistic test of whether the combination is necessary for objective neurophysiological change.

The paper's central move was to disassemble the method, holding everything constant except for which signal components were actually driving the feedback.

What Did They Study?

The study was structured as three tightly matched experiments in healthy participants, each built as a randomized, sham-controlled, double-blind, crossover design. In each experiment, participants completed a verum session and a sham session in counterbalanced order, allowing the authors to test whether brain changes from pre to post differed reliably between the two conditions.

Each experiment isolated one signal recipe.

Study 1 provided only classic frequency-band feedback versus a sham condition. Study 2 delivered only the infra-low frequency feedback versus sham. Study 3 delivered both FB and ILF simultaneously, compared with sham. The primary endpoint was not symptoms or task performance. It was a change in the brain's large-scale connectivity architecture, measured with functional magnetic resonance imaging (fMRI) and analyzed as the functional connectome.

"Same setting, same animation, same electrodes, same timing, different signal ingredients."

How Did They Do It?

All three studies used the same session format. Participants completed a pre-scan, received a single 30-minute neurofeedback block, then completed a post-scan and state questionnaires. The sham condition was implemented by replaying a prerecorded EEG while simultaneously monitoring the participant's real EEG for artifact detection, thereby keeping the feedback plausible and preserving blinding.

Electrode geometry was standardized. The protocol used a two-channel bipolar derivation reconstructed from a 31-channel montage, with right temporal and right parietal sites referenced to the midline. The feedback was delivered as a continuous audiovisual experience: a slow journey through a wooded landscape with music.

The feedback signal was computed as two separable components: FB and ILF.

The FB component extracted power from multiple classic frequency bands and used adaptive thresholds that tracked ongoing fluctuations, with the distance to threshold controlling elements such as music volume and ambient fog.

The ILF component was low-pass filtered to capture the infra-low frequency range, and its power controlled motion speed, brightness, and color saturation. Depending on the study, participants saw only the FB mapping, only the ILF mapping, or both mappings simultaneously.

Across all studies, participants were instructed to observe passively, with no explicit goals or reinforcement cues, a defining feature of implicit neurofeedback in this paper.

The primary analysis used functional connectivity multivariate pattern analysis (fc-MVPA), a voxel-wise, hypothesis-free method that tests whether each voxel's whole-brain connectivity pattern changes as a function of condition. When fc-MVPA identified a significant cluster, the authors then ran post-hoc seed-to-voxel analysis to map which specific regions showed altered coupling with that seed.

What Did They Find?

The headline result was highly discriminative: robust, statistically corrected connectivity effects were observed only when FB and ILF were delivered together. For FB-Only and ILF-Only, the interaction effects did not survive correction. For FB and ILF simultaneously, corrected clusters concentrated in posterior brain regions, including parietal, occipital, and posterior cingulate areas.

"Only the combined signal produced corrected whole-brain connectivity change."

When those clusters were used as seeds, the combined condition showed increased coupling between the posterior midline and parieto-occipital regions and the right dorsolateral prefrontal cortex, and between the posterior cingulate and precuneus regions and the occipital areas. The sham condition did not show comparable suprathreshold connectivity changes in these follow-up maps, strengthening the inference that the effect was attributable to the signal content rather than simply to the experience of watching a compelling animation.

The paper also examined psychophysiology and subjective state as secondary outcomes. On the physiological side, it reported a verum versus sham interaction for heart rate variability (HRV) indexed by the RMSSD only in the combined FB and ILF experiment. The authors interpreted this pattern as broadly consistent with condition-specific effects on parasympathetic influences on the heart, while also emphasizing that mechanisms and links to the connectivity findings remain to be clarified.

Subjective ratings added a realism check on blinding and experience. Participants reported a stronger impression that the feedback "came from me" during verum than during sham, a signal-origin effect, in both the FB-Only and combined FB and ILF protocols.

In ILF-Only, participants reported better well-being during verum than during sham and a greater ability to regain focus after distraction, alongside shifts suggesting changes in arousal state.

Overall, these experiences supported the idea that even a passive protocol can produce condition-differentiated effects that participants partly detect phenomenologically.

Strengths and Limitations

A major strength was design discipline. By running three matched experiments and varying only the feedback signal recipe, the authors reduced a common ambiguity in neurofeedback research in which protocol differences, therapist interaction, and participant expectations are confounded with the signal itself.

Their use of a randomized, sham-controlled, double-blind, crossover design was explicitly designed to isolate neurofeedback-specific effects from nonspecific influences.

The replay-based sham, while monitoring live EEG for artifacts, further strengthened internal validity by preserving the verum's appearance.

The analytic pipeline was another strength. Using fc-MVPA as a hypothesis-free screen, followed by corrected seed-to-voxel analysis, fit a domain where the precise neural targets of these protocols remain uncertain.

The limitations were equally informative. First, this was a single-session experiment in healthy participants; therefore, the work cannot claim clinical benefit or address durability.

Second, fc-MVPA detects multivariate pattern changes and can miss effects that might appear under different connectivity models, which is why the follow-up mapping should be seen as complementary rather than exhaustive.

Third, the physiological findings, including RMSSD effects, were intriguing but not mechanistically resolved, and they were not directly tied to the observed connectivity changes within a single causal chain in this dataset.

Finally, subjective state ratings were valuable for interpretation, but they should be treated as exploratory unless the instrument is established as psychometrically validated for this specific use.

"This is a mechanism paper, not an outcomes paper, and it behaves like one."

What Was the Impact?

The practical impact is that the paper transformed clinical intuition into a testable engineering claim: if ILF neurofeedback is a two-component system, then the components should be independently interrogable, and the combined configuration should be validated. Here, it did.

The combined FB and ILF protocol produced corrected changes in large-scale functional connectivity after a single session, while the isolated components did not, under the same implicit delivery conditions.

This result provides the field with a concrete target for refinement. If future work links these posterior midline and occipito-parietal coupling changes, and their connection to prefrontal control regions, to cognitive or affective outcomes, clinicians and researchers can move from protocol tradition to protocol rationale.

The study also highlighted that implicit designs are not immune to phenomenological differentiation, which is important for interpreting blinding, expectancy, and engagement in neurofeedback trials.

"Disassembling the method is how you learn what to optimize."

Key Takeaways

Under matched conditions, only the combined FB and ILF configuration produced a corrected change in large-scale connectivity, supporting a synergistic rather than an additive claim.
The primary connectivity effects concentrated in the posterior midline and occipito-parietal regions, with increased coupling to prefrontal control areas in follow-up mapping.
A replay-based sham that preserved experience while monitoring live EEG artifacts enhanced blinding in neurofeedback research.
Condition-specific changes in HRV, indexed by RMSSD, were observed only in the combined protocol; however, the physiological mechanisms underlying these changes and their relationship to connectivity require further investigation.
Participant experience differed by condition even in implicit neurofeedback, reinforcing the need to measure state and perceived contingency alongside neurophysiological outcomes.

Glossary

biofeedback: a method that presents real-time physiological signals back to a person to support learning and self-regulation.

classic frequency bands: a set of conventional EEG ranges used to summarize rhythmic brain activity.

crossover design: a study design in which each participant receives multiple conditions, typically in counterbalanced order.

double-blind: a design in which neither participants nor experimenters know condition assignment during data collection.

EEG: an electrophysiological recording of brain activity measured at the scalp.

FB component: the classic frequency-band component computed from the sum signal T4 + P4, extracting power amplitudes from eight EEG bands and using adaptive thresholds; the average distance to threshold modulates music volume and ambient fog. FB-Only: a verum condition where feedback is computed only from EEG power in the classic frequency bands (for example, delta through gamma), with no infra-low-frequency component driving the feedback.

fc-MVPA: a voxel-wise method that tests whether a voxel's whole-brain connectivity pattern changes across conditions.

feedback signal: the real-time, EEG-derived control stream that drives the animation (and music) during neurofeedback, composed of one or both computable components (FB and/or ILF), depending on the study arm.

fMRI: an imaging method that measures brain activity indirectly through blood-oxygen-level-dependent signals.

functional connectivity: a statistical coupling between activity time series in different brain regions.

functional connectome: a whole-brain map of functional connectivity patterns.

ILF component: the infra-low-frequency component computed from the difference signal T4 − P4, low-pass filtered to capture infra-slow cortical potential shifts; total power controls movement speed, brightness, and color saturation of the animation.

ILF-NFB: a neurofeedback approach that includes an infra-low-frequency feedback component, often combined with classic frequency band feedback.

ILF-Only: a verum condition where feedback is computed only from the infra-low-frequency EEG component (very slow fluctuations), with no classic frequency-band power component driving the feedback.

implicit neurofeedback: a protocol in which participants are not instructed to control the feedback intentionally, and instead observe passively.

infra-low frequency: an extremely slow EEG component derived from infra-slow cortical potential shifts.

neurofeedback: a biofeedback method that uses brain-derived signals to provide real-time feedback.

parasympathetic: a branch of the autonomic nervous system associated with restorative regulation and vagal influences on the heart.

perception of signal origin: a subjective rating reflecting how strongly a participant feels their own internal state generates the feedback signal.

randomized: assigned by chance to condition order or group to reduce systematic bias.

RMSSD: a time-domain heart rate variability metric computed as the root mean square of successive differences between interbeat intervals.

seed-to-voxel analysis: a connectivity analysis that tests coupling between a defined seed region and all other voxels in the brain.

sham-controlled: compared against a control condition designed to mimic the experience of treatment without the active signal contingency.

verum: the active neurofeedback condition in which the animation is driven in real time by the participant’s own EEG-derived feedback signal (as opposed to a replayed signal in sham).

Reference

de Matos, N. M. P., Stampfli, P., Seifritz, E., & Brügger, M. (2026). Disassembling infra-low-frequency neurofeedback: A neurophysiological investigation of its feedback components. NeuroImage, 325, 121647. https://doi.org/10.1016/j.neuroimage.2025.121647

About the Author

Fred Shaffer earned his PhD in Psychology from Oklahoma State University. He earned BCIA certifications in Biofeedback and HRV Biofeedback. Fred is an Allen Fellow and Professor of Psychology at Truman State University, where has has taught for 50 years. He is a Biological Psychologist who consults and lectures in heart rate variability biofeedback, Physiological Psychology, and Psychopharmacology. Fred helped to edit Evidence-Based Practice in Biofeedback and Neurofeedback (3rd and 4th eds.) and helped to maintain BCIA's certification programs.