Dynamic pictures, steadier arousal, better neurofeedback signals

A new emerging pilot study examined a deceptively simple question with big consequences for real-time fMRI neurofeedback: how long should we show emotional images if we want the amygdala to stay meaningfully engaged across a training block? The amygdala is often targeted for downregulation in trauma-related symptoms and anxiety, but it’s also famously prone to habituation—a natural fading of physiological and neural response with sustained exposure. If the amygdala “calms down” simply because the stimulus gets old, a neurofeedback display could look like progress even when the brain is merely doing what brains do: adapting.

This matters because neurofeedback—whether with fMRI or EEG—depends on a reliable feedback signal that tracks a person’s internal state in real time. In general terms, neurofeedback is training that provides moment-to-moment information about brain activity so the person can learn self-regulation; biofeedback does the same using bodily signals such as heart rate variability, respiration, or skin conductance. When the training target is emotion circuitry, the quality of the eliciting stimulus (and its timing) is not a cosmetic detail—it becomes part of the “instrument” that shapes learning.

To explore how stimulus timing may influence habituation, the authors compared two ways of presenting aversive images while continuously measuring skin conductance, a peripheral marker closely tied to autonomic arousal and strongly linked with amygdala engagement. Their goal was pragmatic: choose a presentation approach that keeps arousal “alive” long enough to support interpretable neurofeedback blocks.

Methods

The study enrolled healthy adults and used a within-subject design: each participant experienced both a dynamic and a static aversive-image presentation format, with block order randomized to reduce simple sequence effects.

Stimulus timing conditions

Each trial lasted 24 seconds in both conditions:

• Dynamic condition: six aversive images were shown consecutively, each for 4 seconds (six distinct onsets per 24-second trial).

• Static condition: one aversive image remained on screen for the full 24 seconds (one onset per trial).

Participants completed four task blocks total: two dynamic blocks (20 trials each) and two static blocks (21 trials each). The slight difference in trial count was used to keep picture-category exposure balanced across conditions.

Stimulus selection and categories

Aversive pictures were drawn from standardized emotion-elicitation databases (IAPS, GAPED, and NAPS) and filtered to be highly negative and highly arousing. Images were also categorized (e.g., snake/bug, animal, human, social, gore/body, object/scene) to diversify content within trials and across blocks.

Skin conductance as the key outcome signal

Skin conductance was continuously recorded with a BIOPAC system. Electrodes were placed on the pads of the index and middle fingers of the participant’s left hand, and participants were instructed to keep that hand still and refrain from talking during recording. A short breath-holding task was used to confirm that participants showed measurable skin conductance responsiveness.

The authors focused on two common electrodermal indices:

• Skin conductance response (SCR): brief peaks following stimuli.

• Skin conductance level (SCL): a slower, tonic measure of arousal across time.

Data were preprocessed (down-sampling, normalization, detrending, bandpass filtering, smoothing) and segmented so that analysis targeted the image-viewing periods specifically. For SCL time courses, the authors used a generalized additive model to capture the non-linear rise and decay of tonic arousal across the 24-second window for each condition.

Behavioral checks

After each trial, participants rated valence and arousal using a standard self-assessment scale. After the final block, a recognition memory test assessed whether images were remembered better in static or dynamic viewing.

Results

Autonomic arousal over time

The central finding was a difference in how arousal was sustained, not whether participants reacted at all.

Across the 24-second window, tonic arousal (SCL) was higher in the dynamic condition than in the static condition. The shape of the response mattered:

• Static viewing produced a sharper early rise in SCL during the first few seconds, then decayed more rapidly.

• Dynamic viewing produced a more sustained SCL profile, with small increases that aligned with each new image onset.

This pattern is consistent with the idea that novelty within a block can keep the autonomic system engaged, limiting within-trial habituation.

Phasic responses: number of SCR peaks

Despite an intuitive expectation that six images should create more discrete peaks than one image, the number of SCR peaks did not differ meaningfully between conditions. The most plausible explanation offered by the authors was procedural: there were no inter-stimulus or inter-trial intervals, so SCRs likely overlapped in time and did not reliably “reset” between consecutive images.

Anxiety and tonic arousal

Anxiety scores were related to tonic arousal in a condition-specific way. Higher overall anxiety scores correlated with lower SCL during static viewing, while no statistically reliable relationship emerged during dynamic viewing. This suggests that individual differences (and perhaps spontaneous coping strategies) can influence physiological reactivity—especially when the stimulus is held long enough to allow appraisal, distancing, or other regulation efforts to unfold.

Recognition memory

Participants recognized static-condition images better than dynamic-condition images, which is unsurprising: they had 24 seconds with each static image but only 4 seconds per image in the dynamic condition. Practically, this also implies that dynamic presentation may preserve “novelty” across repeated sessions—important when available image sets are limited.

Discussion

If you’ve ever watched a scary movie scene twice, you know the first viewing is a different creature than the second. The nervous system is an efficiency machine: it reduces its response to what it predicts. This pilot study brings that everyday truth into the world of neurofeedback design.

The key practical message is that stimulus timing can create the illusion of neurofeedback success. When a single aversive picture stays on screen for a long block, the amygdala and the autonomic system can naturally “downshift” over seconds—even without any deliberate self-regulation. In a neurofeedback setting where the goal is to train amygdala downregulation, that downshift could be misread as skill acquisition. Dynamic presentations—by refreshing novelty within the block—appear more likely to maintain tonic arousal, making it easier to interpret changes as true self-regulation rather than passive adaptation.

There’s also a quiet but important nuance: dynamic and static formats may recruit different blends of within-trial and between-trial habituation. Static trials encourage within-trial adaptation (the stimulus stays put, so the body relaxes). Dynamic trials reduce within-trial adaptation (new onsets keep arriving), but may change between-trial expectations (participants learn the rhythm of novelty). In real training protocols, this distinction matters because neurofeedback learning is shaped not only by what the brain is doing, but by what the brain expects is about to happen.

From a clinical lens, this has immediate relevance for trauma-related symptoms and anxiety. Exposure-based therapies and emotion-regulation training both live and die by dosing: too little engagement and nothing changes; too much engagement and the person either disengages or becomes overwhelmed. A stimulus format that sustains moderate arousal—without the rapid drop-off seen in prolonged static viewing—could create a more stable “training window” for practicing regulation strategies.

The anxiety finding adds another layer: higher anxiety was associated with lower tonic arousal during static viewing. One interpretation is that longer exposures provide time for spontaneous cognitive strategies—reappraisal, distancing, numbing, distraction—especially in people who live with frequent worry and may have a well-rehearsed coping repertoire. In neurofeedback, this matters because two people can show the same physiological readout for very different reasons: one may be genuinely regulating, another may be checking out.

This is where multimodal measurement becomes valuable. Pairing brain-based feedback (fMRI or EEG) with peripheral measures such as skin conductance can help distinguish true regulation from habituation or disengagement. Skin conductance has higher temporal resolution than fMRI, and it can serve as a running “engagement meter” during blocks. Even in EEG practice, adding electrodermal measures can help tailor session pacing, difficulty, and stimulus design, especially when training targets are emotion-laden.

Finally, the memory results hint at a practical advantage of dynamic presentation across multi-session programs: less explicit recognition may help preserve novelty and reduce the “I’ve seen this one” effect that drains emotional signal over time.

Brendan’s perspective

In a perfect world, neurofeedback would be a clean conversation between a brain signal and a learning system: the brain shifts, the feedback responds, the person learns. In the real world, it’s more like trying to have that conversation in a busy café while someone keeps changing the music. The café noise is not “bad data,” it’s context—and emotional tasks, especially those that use aversive stimuli, are some of the noisiest environments we can choose.

This paper quietly highlights a trap that I see in practice all the time: low arousal can mean three different things.

First, it can mean regulation. The person is present, engaged, and successfully applying skills.

Second, it can mean habituation. The nervous system is simply adapting to a repeated or prolonged stimulus, whether the person is doing anything intentional or not.

Third, it can mean disengagement. The person has checked out, numbed, distracted, or cognitively distanced—not necessarily in a pathological way, but enough that the training target is no longer what we think it is.

The static-picture condition in this study is a beautiful demonstration of how quickly those meanings can blur. A big early spike followed by a fast drop looks like “downregulation” if you only look at the end of the block. But if the drop is driven by habituation, you’ve just rewarded the brain for doing what it would have done anyway. That’s not learning. That’s misattribution.

Why I like adding electrodermal activity to EEG neurofeedback

Electrodermal activity (EDA) is one of the most practical add-ons to EEG work when the training involves anxiety, emotion, vigilance, or anything “threat-adjacent.” It gives you a time-resolved window into sympathetic arousal that often shifts faster than EEG band power trends, and it can flag the moment a person either ramps up or quietly disengages.

In sessions, I’ll often treat EDA as an “engagement dial,” not a success meter. The goal is not to drive it down at all costs. The goal is to titrate arousal into a workable range where learning can happen. If the person’s EDA is flatlining while they report feeling overwhelmed, that’s a clue that they may be dissociating or (especially when not using professional/lab grade equipment), it simply mean that the signal quality is poor. If EDA is climbing modestly and gradually while they report calm competence, that’s not necessarily a problem either: some people experience productive effort as arousal, especially early in training. Follow it, and see where it levels out. 

The practical setup is simple: track tonic level (SCL) across the session and phasic responses (SCRs) around task events, then compare that with what the EEG is doing during the same windows. If you can, add respiration and heart rate variability as well. Not because more channels make you more scientific (who doesn't love even more wires?), but because emotion regulation is a whole-body sport. 

A clinical workflow for “quiet coping” in anxiety

The anxiety finding in this study is the one I’d pin to the wall: higher anxiety scores were linked with lower tonic arousal during static viewing. This is the kind of result that looks counterintuitive until you meet enough anxious people. Many of them have become experts at coping silently. They can dampen physiological reactivity through cognitive control, avoidance, suppression, or a kind of practiced internal distancing. Sometimes that’s adaptive. Sometimes it’s costly. Either way, it means you can’t assume that “low arousal” equals “feeling safe.”

Here’s how I handle this clinically:

1. I begin by mapping the person’s internal experience to their signals. During early sessions, I’ll pause every few minutes and ask for quick subjective ratings: perceived anxiety (0–10), sense of effort (0–10), and sense of connection to the task (0–10). These are not fluffy add-ons. They are calibration points.

2. I watch for mismatches. If someone reports high distress but shows dropping EDA and a drifting EEG pattern, I suspect disengagement or a shift into cognitive avoidance. If they report calm but EDA and frontal tension markers are rising, I suspect unrecognized effort or performance pressure.

3. I adjust the task before I adjust the person. This is a theme I wish more studies made explicit: stimulus design is part of the intervention. If static viewing invites rapid habituation, I can adapt: shorten exposures, add novelty, or rotate stimulus categories and attentional focus. I’d rather tweak the environment than incorrectly reward an unhelpful internal strategy.

Translating this into EEG protocols

This paper is grounded in real-time fMRI logic, but the lesson transfers: don’t confuse signal drift with skill.

When anxiety is the primary concern, I commonly work with a few EEG targets, choosing based on presentation, baseline EEG patterns, and how the person responds during the first sessions.

SMR stabilization (12–15 Hz) at C3, Cz or C4 is my workhorse when the person’s problem is agitation, impulsive reactivity, or sleep fragmentation. The intention is not to sedate; it’s to help the nervous system find a steadier “idle speed.” If I’m pairing this with EDA, I’m looking for higher stability and reduced volatility rather than a monotonic drop. A session where EDA shows fewer sudden spikes and the person reports less hair-trigger reactivity is a win, even if the average level doesn’t plummet.

Posterior alpha enhancement (roughly 8–12 Hz) at POz can be useful when the person is tense, over-monitoring, or stuck in a high-effort attentional stance. But here’s the caution: alpha can rise for two reasons that look identical on a graph. One is relaxed presence. The other is disengagement. EDA helps disambiguate. If alpha increases while EDA remains responsive to task events and the person reports being present, that’s often productive. If alpha increases while EDA flattens and the person reports fogginess or emotional distance, we’ve drifted into a different state.

Frontal midline theta support (around 4–7 Hz) at Fz or FCz, often framed as enhancing focused internal control, can be helpful for anxious rumination when used carefully. I like it most when the person can stay emotionally engaged without getting flooded. Again, EDA gives you the guardrails: if theta training coincides with rising tonic arousal and the person reports strain, you may be asking for “control” in a way that increases threat monitoring.

For trauma-related presentations, I’m cautious with anything that resembles prolonged static exposure early on. This paper supports that instinct. I’ll often use shorter emotional exposures, more variety, and more frequent resets. Sometimes that means alternating brief activation with brief downshift periods rather than trying to ride out a long block. In EEG terms, that might look like brief alpha up-training intervals interleaved with grounding and paced breathing, while monitoring EDA to ensure we’re not drifting into shutdown. You can, in theory and properly following state-dependent implementation, both reward and inhibit posterior alpha for the same person within the same session; we're essentially teaching how to shift in and out of the default mode network. 

Multiple signals, one interpretation

The biggest upgrade you can make to emotion-focused neurofeedback is not a fancier algorithm. It’s a better interpretive stance.

I like to think in triangles: physiology, neurophysiology, and subjective report. If one corner disagrees with the other two, don’t force it to fit. Investigate.

• When EEG shifts look great but the person feels worse, the protocol may be rewarding an unhelpful strategy.

• When EDA drops but the person reports numbness, that may be disengagement, not recovery.

• When the person reports feeling calmer but EDA and EEG suggest rising strain, you may be seeing a “white-knuckle calm,” the kind that collapses outside the office.

This is also where I’m happy to be critical of research designs without being cynical. A study can be methodologically clean and still miss the lived complexity of regulation. Habituation is not a nuisance variable; it’s a core property of the nervous system. If we ignore it, we end up training the easiest-to-get signal change rather than the most meaningful one.

So my clinical takeaway from this paper is straightforward: design tasks that preserve interpretability, use EDA to track engagement, and treat the client’s lived experience as essential data. When those three align, neurofeedback stops being a guessing game and starts looking like what it’s supposed to be: guided learning in a nervous system that is finally getting clear feedback about itself.

Conclusion

This pilot study offers a practical reminder for neurofeedback research and practice: the feedback signal is only as interpretable as the task that generates it. When aversive images are presented dynamically—six brief images within a 24-second block—tonic autonomic arousal stays higher and shows small re-engagements with each new onset. When one aversive image is held on screen for the full block, arousal can spike early and then fade, increasing the risk that habituation will masquerade as successful downregulation.

In real-time fMRI neurofeedback targeting the amygdala, this distinction matters because training often spans long blocks and multiple sessions, precisely the conditions that make habituation likely. The findings also underscore why peripheral measures like skin conductance can be more than “extra data”: they can serve as a time-resolved companion signal that helps interpret whether the brain is regulating, adapting, or disengaging.

When we design neurofeedback protocols with stimulus timing in mind, we’re not just improving experimental neatness—we’re improving the odds that what looks like learning actually is learning.

References

Gura, H., Wiener, E., Beynel, L., Luber, B., & Hollingsworth Lisanby, S. (2026). The impact of stimulus presentation on skin conductance and potential implications for neurofeedback studies. Applied Psychophysiology and Biofeedback. https://doi.org/10.1007/s10484-025-09757-3