Disassembling Infra-Low Neurofeedback

This new emerging research with novel insights, led by de Matos and colleagues, asks a deceptively simple question: when we do infra-low-frequency neurofeedback, which part is actually doing the heavy lifting – the classic EEG frequency bands, the ultra-slow infra-low activity, or the combination of both? Using three carefully controlled fMRI studies, the authors systematically pull apart the components of ILF neurofeedback to see how each one reshapes brain connectivity and physiology.

Neurofeedback is a form of biofeedback that uses real-time brain signals to help people learn to self-regulate neural activity, usually by turning EEG or fMRI signals into simple visual or auditory feedback. Biofeedback more broadly does the same thing for bodily signals like heart rate, breathing, or skin conductance. Both approaches sit at the intersection of neuroscience and learning theory, where the brain essentially becomes both the student and the teacher.

Infra-low-frequency neurofeedback is a relatively young but fast-growing variant of EEG neurofeedback. Instead of asking people to consciously push bars up or down, ILF training is usually implicit: the client simply watches an animation whose movement or sound is subtly driven by very slow cortical potentials and, in many protocols, by classic EEG bands at the same time. Clinically, ILF is often described as a “global regulation” method that calms stormy networks, but the exact mechanisms have remained murky.

This study takes a major step toward clarifying those mechanisms. By combining high-quality resting-state fMRI, advanced connectivity analyses, and a strict double-blind sham design, the authors compare three otherwise identical protocols that differ only in which EEG components drive the feedback. The result is a rare, mechanistic look under the hood of ILF neurofeedback.

Methods

The work is actually three parallel, randomized, sham-controlled, double-blind crossover studies, each with 40 healthy adults who completed two sessions: one verum neurofeedback session and one sham session, at least two weeks apart. Across all three studies, the structure was identical: preparation and questionnaires, a pre-intervention resting-state fMRI scan with psychophysiology, a 30-minute neurofeedback block outside the scanner, and then a post-intervention resting-state fMRI scan.

Participants were screened to be physically and mentally healthy, medication-free, and MRI-compatible. Each study used its own independent cohort, so no one took part in more than one protocol.

Neurofeedback was delivered with a 34-channel EEG cap, but feedback was derived from a two-channel bipolar montage reconstructed from electrodes T4, P4, and Cz. All protocols used the same visual environment: a slow, immersive animation called “Dreamscapes – Wilderness,” with music and subtle visual changes. Importantly, all three protocols were fully implicit: participants were told to simply watch and not try to control anything.

The key manipulation was which EEG components drove the feedback:

• FB-Only (Study 1): The software summed activity at T4 and P4 and extracted power from eight classic frequency bands between 2 and 40 Hz (delta through low gamma). Adaptive thresholds for each band were continuously updated to cover 95 percent of ongoing fluctuations. The average distance to these thresholds modulated music loudness and ambient fog in the animation. No infra-low component was used.

• ILF-Only (Study 2): The infra-low component was derived from the difference signal T4 minus P4, low-pass filtered to capture ultra-slow potential shifts below the conventional EEG range. The total power of this infra-slow signal controlled movement speed, brightness, and colour saturation of the scene. No classic frequency-band feedback was added.

• FB&ILF combined (Study 3): Both components were active simultaneously. Frequency-band activity modulated music volume and fog, while the ILF component controlled speed, brightness, and saturation. This mirrors clinical ILF protocols where FB and ILF are typically presented together.

Sham sessions replayed prerecorded EEG from another participant through the same software, while the system still analysed the real EEG in the background to inject realistic artefacts, preserving blinding. Verum–sham order was randomized by the software and hidden from both participants and experimenters.

The primary outcome across all three studies was functional connectivity multivariate pattern analysis (fc-MVPA) of resting-state fMRI. For each voxel, the authors computed its correlation profile with the rest of the brain, reduced the dimensionality to dominant eigenpatterns, and then tested for changes across conditions. The critical contrast was the interaction [Post − Pre Verum] versus [Post − Pre Sham], asking where connectivity changed more after real neurofeedback than after sham. Strict false discovery rate (FDR) correction was applied at both voxel and cluster level.

Significant clusters from this multivariate analysis were then used as seeds in post-hoc seed-to-voxel analyses, again with conservative correction (threshold-free cluster enhancement with Bonferroni-adjusted family-wise error). Secondary outcomes included heart rate, heart rate variability (RMSSD), respiration, and a state questionnaire probing subjective experience of the feedback.

Results

The headline result is straightforward but powerful: only the combined FB&ILF protocol produced robust, statistically corrected changes in whole-brain functional connectivity. Neither FB-Only nor ILF-Only showed interaction effects that survived FDR correction in the main fc-MVPA analysis, although more lenient thresholds suggested some weaker patterns.

For the FB&ILF study, the interaction contrast revealed three significant clusters:

• A right posterior supramarginal gyrus cluster in the inferior parietal lobule.

• A cluster centred in the precuneus with extensions into left supracalcarine and cuneal occipital cortices.

• A cluster in the posterior cingulate gyrus with small contributions from the left precentral gyrus and adjacent precuneus.

These regions sit along the posterior midline and parieto-occipital junction – classic hubs of the default mode and visual networks.

When these clusters were used as seeds, seed-to-voxel analyses showed that, for verum FB&ILF only, connectivity increased between:

• The precuneus/occipito-parietal cluster and the right dorsolateral prefrontal cortex (frontal pole).

• The posterior cingulate seed and a broad swath of medial parieto-occipital cortex, including precuneus, cuneal, intracalcarine, supracalcarine cortices, lingual gyri, and cerebellar vermis.

• The same posterior midline seed and additional precuneus and lateral occipital regions extending to the occipital pole.

No significant connectivity changes were observed for the sham condition at the corrected level. Cluster 1, in the right posterior supramarginal gyrus, did not yield suprathreshold seed-to-voxel results under the very strict correction, although more liberal thresholds suggested more diffuse effects.

Psychophysiological data painted a subtler picture. Heart rate decreased over time in both verum and sham sessions across all three studies, likely reflecting general habituation to the scanner. Respiratory cycle duration showed a condition-by-time interaction across studies, with somewhat greater changes in sham, but this effect was sensitive to a small number of outliers.

Heart rate variability, indexed by RMSSD, showed a significant verum–sham difference only in the FB&ILF study: increases in RMSSD were larger after sham than after verum. A sensitivity analysis excluding participants with extreme respiratory changes preserved this RMSSD effect while removing the respiration interaction, suggesting that the HRV result was not simply an artefact of breathing patterns.

Questionnaires revealed that participants in FB-Only and FB&ILF conditions felt more strongly that the feedback signal “came from them” and was more controllable during verum than sham. ILF-Only verum sessions were associated with better well-being and slightly easier refocusing after distraction, but also with more sleepiness and a small shift toward nervousness. Interestingly, the combined FB&ILF protocol led to lower perceived activity during verum, consistent with a more settled internal state, even though autonomic markers did not show straightforward “more relaxation with verum” patterns.

Discussion

Taken together, these findings suggest that the classic ILF clinical recipe – combining infra-low and frequency-band components in a single, implicit protocol – is not just a historical accident. When both signals drive feedback simultaneously, the brain’s large-scale connectivity patterns shift in a way that is stronger, more widespread, and more statistically robust than when either component is used alone.

The most striking changes sit at the crossroads of the default mode, visual, and executive control networks. Increased coupling between precuneus/occipito-parietal cortex and right dorsolateral prefrontal cortex hints at enhanced communication between internal, self-referential processing hubs and a key region for top-down control, working memory, and cognitive flexibility. In parallel, the posterior cingulate’s strengthened links with occipital and cerebellar regions suggest a re-tuning of how internally generated states are integrated with incoming sensory information and sensorimotor prediction.

Clinically, this is intriguing. Many people who seek neurofeedback – whether for anxiety, trauma, ADHD, or burnout – struggle with exactly this dance between inner experience and executive control: too much rumination, too little flexible shifting, or difficulty staying grounded in the present. The connectivity pattern emerging here looks like the brain briefly reorganizing its backstage crew: posterior hubs still generate and integrate internal models, but they appear to be in closer conversation with a frontal supervisor.

From a practical standpoint, this work adds weight to the idea that ILF protocols may be most potent when paired with at least some form of frequency-band engagement. Even though the FB component here did not explicitly train individual bands like SMR or high beta, its presence appears to change how the infra-slow system talks to the rest of the brain. One possible interpretation is that ILF engages a very slow regulatory backbone, while frequency-band dynamics provide richer, higher-frequency variation for the system to “learn from,” leading to more pronounced network-level plasticity.

For people considering neurofeedback, one key message is that meaningful brain changes can occur even in a single implicit session, without effortful strategy use or consciously trying to make anything happen. You do not have to be good at visual imagery, meditation, or “concentrating hard” for your brain to respond. In this paradigm, the brain receives a gentle, continuous mirror of its own infra-slow and classic-band dynamics and appears to reorganise its connectivity within minutes.

For clinicians who refer to neurofeedback, the study is also reassuring in terms of methodological rigour. The protocols were double-blind, sham-controlled, and analysed with modern connectivity statistics. Importantly, the sham condition was not just a flat-line placebo; it replayed realistic EEG with artefacts, making it genuinely difficult for participants or experimenters to guess the condition. The fact that FB&ILF still produced unique connectivity patterns over and above this rich control condition suggests that the signal content itself matters, not just the relaxation, expectation, or “techno-ritual” of sitting in front of a screen.

For neurofeedback practitioners, the details are particularly juicy. The montage (T4–P4) targets a right posterior temporal–parietal axis often used clinically for regulation of arousal, sensory processing, and emotional reactivity. The main connectivity shifts showed up in neighbouring parietal and midline hubs and their links to right dorsolateral prefrontal cortex. This aligns with clinical impressions that T4–P4 ILF work tends to affect self-regulation, sensory integration, and cognitive flexibility rather than narrow, task-specific functions.

The interpretive layer, however, calls for caution. The sample consisted of healthy young adults undergoing only a single 30-minute session per condition. Clinical ILF work typically involves 20–40 sessions, progressive adjustment of ILF frequencies, and flexible montages based on symptom patterns. It is entirely possible that ILF-Only or FB-Only protocols would show clearer neurophysiological effects over repeated sessions or in populations with more room for improvement. The absence of strong single-session effects for those conditions in healthy brains should not be overgeneralised to mean they are ineffective clinically.

Still, the pattern fits a broader theme in neurofeedback research: multimodal or composite protocols often tap into network-level regulation more effectively than highly narrow, single-parameter approaches. The combined FB&ILF protocol may help synchronise slow regulatory scaffolding with faster functional rhythms, nudging the system toward a more integrated, flexible mode of operation.

The autonomic findings are more ambiguous. Increased HRV after sham compared to verum in the combined study is counterintuitive if one equates “good neurofeedback” with “more parasympathetic tone.” But vagal regulation is context-dependent, and higher HRV is not always “better” in every short-term experimental setting. Here, the effect might reflect different cognitive or attentional states during the post-intervention scan rather than simple relaxation. The sensitivity analysis suggests that these HRV shifts are not just artefacts of breathing, but they remain difficult to map onto a clean narrative.

Broadening the lens to other research, these results echo prior ILF work showing changes in large-scale networks, including salience and visual systems, after ILF-based feedback that combined infra-low and frequency-band components. They also resonate with explicit neurofeedback studies where increased coupling between prefrontal control regions and sensory or limbic areas accompanies improved self-regulation. The twist here is that ILF operates implicitly: even without explicit strategies or reward cues, the brain still appears to reconfigure its connectivity in meaningful ways when its own signals are mirrored back in these composite protocols.

Brendan’s perspective

I’ll confess something up front: I’ve never been an infra-low-frequency groupie.

Part of that is temperament. In clinical work I like signals I can see, define, and defend in a journal club without needing a minor act of faith. Classic EEG frequencies, slow cortical potentials, HRV, respiration – they’re messy, but at least we have reasonably clear models of what they reflect and how we’re measuring them. ILF, by contrast, has often felt like the mysterious cousin who shows up at family dinners with big stories and very little documentation.

This paper doesn’t make me an ILF convert overnight, but it does nudge me from “Really?” toward “Okay, tell me more.” And that shift is worth unpacking.

First, the measurement issue. Ultra-slow activity is notoriously hard to capture: you need rock-solid DC stability, excellent impedance control, and a ruthless stance toward artefacts (sweat, movement, drifts, you name it). In many day-to-day clinical setups, ILF is derived with equipment and montages that would make a hardcore electrophysiologist twitch. That’s one of the reasons I’ve been cautious: if we’re going to build an intervention around a signal, I want to be reasonably sure we’re not just conditioning slow artefacts and calling it “deep regulation.”

What I appreciate in this study is that the authors don’t solve that problem completely – but they do show something reproducible happens when ILF and classic bands are combined. The fact that the combined FB&ILF protocol, and only that protocol, changes connectivity between posterior hubs and right dorsolateral prefrontal cortex after a single session suggests we’re not just staring at noise. Whatever exact biophysics hides under the ILF hood, the composite signal appears to be meaningful enough for the brain to reorganise around it.

Then there’s the timing issue. I’ve also been sceptical about how you can meaningfully condition activity that slow in anything like “real time.” It’s a bit like trying to teach someone to dance by giving them feedback on their average posture over the last minute. At some point, the loop feels too sluggish to support precise learning. That’s part of why fMRI neurofeedback, which also works with slow-ish signals, tends to require a lot of averaging and patience.

Here, the clever move is that ILF isn’t working alone. The combined protocol lets ILF ride along with a richer, faster landscape of 2–40 Hz activity. My working hypothesis – and it is very much a hypothesis – is that the infra-slow component may be setting the baseline mood of the system, while the faster bands carry the moment-to-moment variations that make reinforcement and learning feasible. Think of ILF as the tide and the classic bands as the waves you actually surf on.

How does this translate into clinical practice for me?

If I’m honest, this paper doesn’t suddenly make me want to throw out tried-and-true beta, SMR, alpha, or SCP-based protocols and replace everything with ILF. In the ADHD, anxiety, trauma, and mood cases I see, I still lean heavily on better-characterised protocols: for example:

• SMR (12–15 Hz) at C3/C4 for behavioural inhibition and sensorimotor stability.

• Low-beta (15–18 Hz) frontally for sustained attention and cognitive endurance.

• Alpha up-training (8–12 Hz) parietally or occipitally for anxiety and hyperarousal.

• Theta/beta ratio work in carefully selected cases where the qEEG and clinical picture actually support it.

These are protocols where the link between what we train, how we measure it, and what changes we expect behaviourally is at least somewhat constrained by decades of work.

Where this paper does move the needle for me is in how I think about adding ILF as a layer rather than a standalone hero. If I were to integrate it based on these findings, it would probably look something like this:

• Use a T4–P4 or neighbouring montage (I'm allergic to bipolar montages for training) for combined ILF+FB when the main issue is global regulation and integration – people who feel “out of sync” with themselves, with lots of internal chatter and difficulty shifting gears.

• Keep the ILF component relatively conservative and stable in the early sessions, while paying careful attention to clinical responses (sleep, headaches, emotional lability, sensory sensitivity).

• Let the classic frequency component carry more of the explicit weight: supporting SMR and low beta stability, gently discouraging excessive high beta or low-frequency slowing where appropriate.

• Always cross-check with a decent-quality qEEG and symptom tracking, so any ILF-related decisions are made in a broader context rather than on dogma.

I would still be wary of positioning ILF as the core of a neurofeedback practice, especially in the absence of robust measurement infrastructure. But I’m more open to the idea that, when carefully combined with better-understood bands, it might help nudge large-scale networks into a more flexible configuration – exactly what this study suggests with the posterior–prefrontal connectivity shifts.

This also feeds into a bigger point that I care a lot about: the gap between research designs and real clinical work.

In the lab, you freeze almost everything: one montage, one protocol, one session length, minimal interaction. That’s necessary to make clean inferences, but it’s also the exact opposite of how good neurofeedback is typically done day to day. In the clinic, we individualise relentlessly: we tweak frequencies, move electrodes, adjust thresholds, integrate psychotherapy, and respond to subtle shifts in the person’s life between sessions.

So when a tightly controlled, one-off ILF+FB session in healthy adults produces measurable connectivity changes, I don’t interpret that as “ILF is the answer.” I read it more as “There’s something here worth taking seriously – but we still need to earn our certainty.”

If you’re a practitioner who loves ILF, this paper gives you a more respectable story to tell about network-level effects, but it doesn’t absolve us from doing careful assessment, monitoring, and critical thinking. If you’re a practitioner who, like me, has been sceptical, it’s at least a reminder not to dismiss the method outright simply because the mechanisms are uncomfortable or the marketing has sometimes outpaced the data.

And if you’re a client or potential client reading this, the takeaway is simpler: neurofeedback isn’t magic, and not all protocols are created equal, but your brain can and does respond to being gently mirrored – sometimes even when the mirror is built from signals we’re still learning to understand. My bias is to start with approaches where we have clearer measurement and theory, and then consider ILF as an optional layer when the clinical picture and available equipment make it reasonable.

In that sense, this paper might be the first step in convincing me that ILF deserves a place at the table – not as the mysterious guru of neurofeedback, but as one more tool we can use thoughtfully, cautiously, and in combination with the more grounded protocols that already have a solid track record.

Conclusion

This trio of studies offers a rare, mechanistic look at infra-low-frequency neurofeedback stripped down to its core components. In healthy adults undergoing a single implicit training session, classic frequency-band feedback or ILF feedback alone did not produce strong, corrected changes in brain connectivity. In contrast, the combined FB&ILF protocol reliably reshaped connectivity between posterior midline and parieto-occipital hubs and the right dorsolateral prefrontal cortex, with no comparable effects in sham.

The findings suggest that ILF neurofeedback may be most powerful when infra-slow regulatory dynamics and faster oscillatory activity are engaged together, nudging the brain’s networks toward a more integrated and flexible configuration. While single-session autonomic changes were mixed and the work does not directly address clinical outcomes, it provides an important foundation for understanding how ILF protocols might exert their effects. For clinicians and clients alike, the key take-home is encouraging: even a quiet, implicit session of combined ILF and frequency-band feedback can begin to reshape how different parts of the brain talk to each other, opening the door to deeper change over time.

References

de Matos, N. M. P., Stämpfli, P., Seifritz, E., & Brügger, M. (2025). Disassembling infra-low-frequency neurofeedback: A neurophysiological investigation of its feedback components. NeuroImage, 121647. https://doi.org/10.1016/j.neuroimage.2025.121647