The Impact of Stimulus Presentation on Skin Conductance and Potential Implications for Neurofeedback Studies

Abstract

Real-time fMRI neurofeedback enables individuals to learn self-regulation of specific brain regions, such as the amygdala, which is often targeted for downregulation in trauma-related disorders. However, repeated exposure to emotional stimuli during lengthy sessions may lead to habituation, with amygdala activity naturally decreasing over time independent of self-regulation success. Presentation rate may influence this effect––however, empirical evidence remains limited with that rate varying widely across studies (ranging from three pictures every 4.5 s to one picture for 20 s). Therefore, prior to conducting an intended real-time fMRI neurofeedback study, we conducted this pilot to evaluate two types of presentation rate. We compared two aversive image presentation procedures and used skin conductance as an index of amygdala activation to identify an optimal presentation rate for a future neurofeedback study. The final sample included 13 healthy adults, who viewed blocks of six aversive images every four seconds (labeled “dynamic”) and blocks of one aversive image for 24 s (labeled “static”) while skin conductance was continuously recorded. Before the task, participants completed questionnaires assessing stress exposure and anxiety to explore their relationship with skin conductance. Skin conductance level (SCL), a measure of longer-term tonic arousal, remained higher in the dynamic condition than in the static condition. Although the static condition elicited a sharper initial peak, the response decayed more rapidly, whereas the dynamic condition yielded sustained SCL with small increases following each image onset. Further, we found that higher anxiety scores, as defined with the State-Trait Anxiety Inventory for Adults, correlated with lower SCL. These findings suggest that dynamic image presentations may mitigate amygdala habituation more effectively than static presentations, highlighting stimulus timing as a key factor for optimizing protocols such as neurofeedback.

Impact statement

Real-time functional magnetic resonance imaging (fMRI) neurofeedback holds promise for treating psychiatric disorders by enabling self-regulation of emotion-related brain regions. However, its efficacy may be constrained by habituation of target regions after repeated stimulus exposure. This work provides guidance on how aversive visual stimuli can be presented to minimize amygdala habituation, informing future neurofeedback approaches aimed at reducing fear and anxiety symptoms.

Introduction

Real-time fMRI neurofeedback (rt-fMRI-NF) is an endogenous neuromodulation technique that enables individuals to self-regulate activity in specific brain regions and could play a key role in investigating and modulating neural circuits. For example, the amygdala is a key region of emotion processing, and aberrant amygdala activity plays a central role in the pathophysiology of several psychiatric conditions, particularly those involving affective dysregulation such as post-traumatic stress disorder (PTSD) (Wolf & Herringa, 2016), bipolar disorder (Bertocci et al., 2014), generalized anxiety disorder (Fitzgerald et al., 2017), and irritability (Lee et al., 2023). rt-fMRI-NF has been demonstrated to be effective in modulating amygdala activity and improving clinical symptoms for patients with depression (Young et al., 2014) and PTSD (Nicholson et al., 2017).

While advances in rt-fMRI-NF are promising, variation in the stimuli selected and presentation methods across studies makes it difficult to compare results (Goldway et al., 2022). Images are commonly used to elicit emotions for rt-fMRI-NF paradigms and are often obtained from Standardized Emotion Elicitation Databases (SEEDs) (Branco et al., 2023) such as the International Affective Picture System (IAPS) (Lang et al., 1997). However, even within the subset of studies using visual stimuli, the way images are presented varies widely. Some studies present a single static image for an extended duration ranging from 18 (Paret et al., 2018) to 46 s (Paret et al., 2014), while asking participants to regulate their brain activity. In contrast, other studies use a more dynamic approach, using consecutive image presentation (e.g., four images for 1.5 s each in Johnston et al., 2010; three images for 4.5 s each in Sarkheil et al., 2015).

The variation in stimuli presentation has sparked discussions about the most effective methodology in the field, with particular concern about amygdala habituation (Barreiros et al., 2019). Specifically, prolonged exposure to a single stimulus may reduce amygdala reactivity over time and thereby confound interpretation of volitional regulation during neurofeedback.

While in many behavioral paradigms habituation refers to a decreased physiological response to repeated stimuli, we refer to habituation as a decrease in the physiological response throughout image presentation, regardless of whether the images are repeated (a definition used in Breiter et al., 1996). Amygdala habituation is well documented in neuroimaging: both fMRI and PET studies report decreased amygdala activation across repeated trials of aversive stimuli (e.g., Fischer et al., 2000; Wright et al., 2001). However, the impact of presentation format and duration on habituation has not been empirically tested within the context of rt-fMRI-NF paradigms, and no consensus exists regarding optimal stimulus presentation.

To optimize stimulus presentation and minimize habituation, we sought to use skin conductance, a peripheral measure with high temporal resolution and established links to amygdala function.

Skin conductance, which measures the skin’s electrical conductivity influenced by sweat production, serves as a physiological marker of autonomic nervous system activity, reflecting changes in the skin’s properties in response to emotional or arousing stimuli (e.g., Lang et al., 1993; Van Dooren et al., 2012). Two common indices are the skin conductance response (SCR), a phasic and transient increase occurring in reaction to discrete stimuli, and the skin conductance level (SCL), a tonic measure representing the overall level of conductance across time (Braithwaite et al., 2013).

Both SCR and SCL have been used to provide useful information for studying habituation, the gradual reduction of physiological response with repeated stimulus exposure. For example, in the auditory domain, early work using repeated loud tones demonstrated that SCR decreased more at the beginning of the sequence of stimuli than later (Epstein et al., 1978; Lykken et al., 1988; Martin-Soelch et al., 2006). Importantly, the rate of auditory habituation has been used to distinguish clinical populations from healthy individuals. For example, slower habituation has been observed in individuals with depression (Lader & Wing, 1969), chronic anxiety (Raskin, 1975), and in those with lower psychological resilience (Walker et al., 2019) than in healthy controls. Habituation has also been observed in misophonia patients, measured as a decrease in SCL over the course of multiple runs of misophonic, neutral, and aversive sound stimuli, demonstrating this phenomenon is robust across populations and stimulus types (Neacsiu et al., 2024).

Habituation to visual stimuli has also been documented. Bradley et al. (1993) reported rapid habituation of SCR to repeated emotional pictures, with responses reemerging when novel images were introduced. This effect was replicated by Houtveen et al. (2001), who found reduced responses during second exposures to images. However, in a subsequent study, Bradley et al. (1996) did not replicate this habituation effect, suggesting that visual habituation may be complex and context dependent. Habituation of SCRs has also been reported over blocks of masked fear word presentation, with even stronger habituation to cues that elicited larger initial responses, i.e. threat words (van den Hout et al., 2000). The same is true with SCL measures, with SCL habituation being reported across trials of masked phobic images (Siegel et al., 2018).

Skin Conductance & Amygdala

Skin conductance reflects excitatory influences from the amygdala (Boucsein, 1992; Dawson et al., 2017), and both SCR and SCL have been shown to correlate with neuroimaging measures of amygdala activity. For example, one conditioned shock study using fMRI demonstrated that higher left amygdala activation is associated with higher average SCL across a block of threat versus safe conditions (Phelps et al., 2001), and another study demonstrated that significant amygdala activation occurs only when a measurable SCR is elicited (Cheng et al., 2006). Similarly, Labar et al. (1998) reported that the magnitude of conditional SCRs was correlated with the proportion of active voxels in the amygdala during an emotional task. Together, these findings suggest that skin conductance, measured through both SCL and SCR, provide reliable peripheral proxy measures for amygdala activation in response to aversive stimuli.

There is limited work examining how the presentation method of aversive images influences skin conductance. Codispoti et al. (2001) compared short (500ms) and sustained (6s) picture presentations and found that short presentations elicited a similar number of SCRs as longer presentations. However, this study focused on the initial phasic response and did not assess skin conductance over extended periods (e.g., 18–46 s), which better translate to neurofeedback block duration. Smith et al. (2005) investigated the cumulative effect of sustained stimulus presentation by presenting blocks of 12 neutral or aversive pictures for 6 s each. They reported a greater number of SCRs for unpleasant stimuli overall and observed a slight decrease in the number of SCRs over time, suggesting some habituation. However, their design did not include any long-duration presentations, leaving open the question of whether habituation might have been greater with extended exposure.

To address this gap, we used a within-subjects design in which participants were each presented with both static (single image for a long duration) and dynamic (six consecutive images, each presented for shorter durations) presentation rates, in a randomized order, while measuring skin conductance, with the goal of informing stimulus design for future real-time fMRI neurofeedback studies. We hypothesized that a dynamic presentation would produce more sustained responses than a static presentation. Specifically, we expected to observe a greater number of distinct SCRs for the dynamic versus static presentation; and a progressive decrease in SCL during static presentations that would remain sustained during dynamic presentations.

Methods

Subjects

All participants signed an informed consent form, and the protocol was approved by the National Institutes of Health Intramural Institutional Review Board under protocol 18-M-0015. This study served as a pilot for Substudy 3 of a larger clinical trial registered at clinicaltrials.gov (NCT03351764). The present results, analyses, and interpretations are not directly included in the preregistration.

Of 16 healthy volunteers enrolled (8 female, average 27.31 ± 5.41 years old), and 15 completed this study (7 female, average 27.53 ± 5.53 years old). Sample size was informed based on a comparison to sample sizes in similar studies (Cheng et al., 2006; LaBar et al., 1998; Phelps et al., 2001), as well as feasibility constraints involving performing a pilot study. The one non-completer discontinued after one block of our task, as the images made them uncomfortable. All subjects were native English speakers. Exclusion criteria included pregnancy or current breastfeeding, history of any DSM-5 disorder previously classified under Axis I, history of epilepsy, stroke, brain surgery, or head trauma, increased risk of seizure, history of drug or alcohol abuse, presence of ferromagnetic metal in the body, history of an unstable or serious medical or neurological disorder, a positive test for HIV, and hearing loss. All participants provided written informed consent before starting the procedure.

The present results, analyses, and interpretations are not directly included in the preregistration. All data (skin conductance level, skin conductance response and hit rate from the recognition memory task) are publicly available on Open Science Framework (https://osf.io/nr7dw/).

Questionnaires

After consenting to the protocol, participants completed a set of stress and anxiety questionnaires, including a custom-made stress exposure questionnaire to assess their amount of exposure to violent media (measured by the questions “Do you watch horror movies?” and “If you watch horror movies, what types do you watch?”) as well as the image categories they find the most distressing or disturbing (e.g. car crash; sexual harassment/domestic violence; shootings/mass casualty incidents/war; natural disasters; blood; medically-related scenes; bugs/snakes; animals/animal cruelty; violence); the State-Trait Anxiety Inventory for Adults (STAI-AD, Spielberger 1971), an instrument meant to measure anxiety which distinguishes between temporary anxiety, or “state anxiety” (S-Anxiety), and long-standing patterns of anxiety, or “trait anxiety” (T-Anxiety); and the Traumatic Life Events Questionnaire (TLEQ, Kubany et al., 2000) to identify the number of traumatic events that may affect a subject’s emotional state and quality of life.

Experimental Session

All sessions were conducted in the same private experimental room to ensure consistent environmental conditions across participants, including sound level, temperature, and humidity. Single-use skin conductance electrodes were attached to the participant’s left hand (see Skin Conductance Recording). Participants were instructed to: (1) keep their left hand still, resting palm-up on the table in front of them, and (2) refrain from speaking during the recording unless they needed to stop the experiment. Adherence to these instructions was verified by continuous observation by experimenters in the experimental room.

We allowed approximately 10 min for electrodes to stabilize before beginning recording, during which we explained the main experiment in detail and answered any questions participants may have had. Participants first performed a breath-holding task in which they were instructed to keep their hands very still and their eyes closed for one minute before taking a deep breath in, holding their breath for three seconds, and breathing out to test whether a subject had robust skin conductance responses (Braithwaite et al., 2013). Participants were then familiarized with the task via one static and one dynamic presentation trial. After the practice trials participants verbally confirmed their understanding then performed four blocks of the task: two blocks with dynamic picture presentations (20 trials/block) and two blocks with static picture presentations (21 trials/block). Block order was randomized to minimize order effect (see Fig. 1). Difference in trial number between static and dynamic blocks was to ensure that each picture category was presented the same number of times, despite the total number of images presented being different. In each trial, aversive images (see Aversive picture selection) were presented for 24 s: six images, each displayed for four seconds in a dynamic trial, versus a single image for the full 24 s in a static trial. At the end of each trial, participants were asked to rate their emotional experience using the Self-Assessment Manikin (Bradley & Lang, 1994), and to rate on a scale from 1 to 9 on two dimensions: valence (the pleasantness of an emotion, ranging from “Very Negative” to “Very Positive”), and arousal (the intensity of the emotional response, ranging from “Relaxed” to “Aroused”).

Fig. 1.

Task design

At the end of the last block of the task, subjects completed a recognition task, allowing us to assess which type of stimulus presentation led to better recognition. The memory task consisted of 48 images, 50% of which were new to the participants, and the other 50% had been previously presented during the task blocks. As each image appeared on the screen, participants were asked, “Have you seen this picture before?” and prompted to provide one of four responses: “Yes, definitely,” “Yes, I think so,” “No, I don’t think so,” and “Definitely no.”

Aversive Picture Selection

Images from the International Affective Pictures System (IAPS) (Lang et al., 2008), the Geneva Affective Picture Database (GAPED) (Dan-Glauser & Scherer, 2011), and the Necki Affective Picture System (NAPS) (Balsamo et al., 2020) were compiled and filtered based on the valence and arousal ratings published in their respective validation studies. We selected highly negative (valence < 3.5 for IAPS & NAPS; < 35 for GAPED) and highly arousing (arousal > 5.5 for IAPS & NAPS; > 55 for GAPED) images. Images were manually categorized into specific categories (snake/bug [mostly snakes and spiders in nature], animal [generally mammals experiencing harm or abuse], human [individuals], social [humans in groups], gore/body, object/scene) by one experimenter, and two additional researchers confirmed the categorization through visual inspection. Due to the limited availability of validated image sources and images within each category and recognizing that different categories may elicit variable emotional responses across participants, we opted to present one image from each category in a random order in each dynamic block. In the static blocks, an equal number of images from each category were shown, as described below. 12 images from the IAPS were excluded despite meeting the valence and arousal criteria, as they were deemed too extreme and unethical to display (Image IDs: 9800, 9810, 2730, 9006, 9321, 9322, 9325, 9326, 9301, 9302, 9320, 4664.2).

Skin Conductance Recording

Skin conductance was recorded with the BIOPAC MP-160 system (BIOPAC Systems, Inc., Goleta, CA) and AcqKnowledge 4.1 software, with a sampling rate of 2000 Hz. The task was presented using Psychopy (Peirce et al., 2019), which was programmed to send event markers to the BIOPAC system at the onset of each image and each rating period. Skin conductance was recorded continuously throughout the session. Disposable dry electrodes (BIOPAC Part # EL509) with compatible isotonic electrode gel designed for skin conductance measures (BIOPAC Part # GEL101A) were attached to the pads of participants’ middle and index fingers of their left hands.

Skin Conductance Data Pre-processing

After data acquisition, we exported our recordings as MATLAB files using the acq2mat.py script from MathWorks. Skin conductance preprocessing involved normalization, detrending, and noise filtering through the following steps: First, data were down-sampled by a factor of 20, resulting in a final sampling rate of 100 Hz. The down-sampled skin conductance voltage signal was then Z-scored by dividing the signal value by its standard deviation and subtracting the mean to center the signal around 0, across the entire recording session. A 4th -degree polynomial was fitted to the signal across the entire recording using least squares regression and subtraction from the signal was used to remove the linear trend. The resulting detrended signal was then bandpass filtered using a Butterworth filter with cut-off frequencies of 0.05 Hz and 5 Hz to remove slow drift and high-frequency noise. Finally, the filtered signal was smoothed using a moving average with a 100-sample (1s) window to suppress rapid, non-physiological fluctuations in the signal (see Fig. 2).

Fig. 2.

Single subject skin conductance waveform. The top waveform is raw data from a participant who underwent two blocks of dynamic presentation then two blocks of static, collected at 2000 samples/second. The middle waveform shows the same participant’s data after all preprocessing steps, including downsampling by a factor of 20 and detrending the signal. The bottom waveform shows data organized into image presentations, with red lines representing the onset of each dynamic image and green lines representing the onset of each static image

As data were continuously recorded throughout the experiment task, which included both image viewing and rating periods, a custom MATLAB script was used to extract the timepoints corresponding specifically to image presentations. Skin conductance data collected during rating periods were excluded based on event markers stored in the recordings. For the breath-holding task, a separate script was developed to compute the mean and maximum SCL during the breath-hold phase, the start of which was indicated with an event marker on AcqKnowledge.

Statistical Analysis

We analyzed two measures of skin conductance to determine which stimulus presentation rate (static or dynamic) leads to the least habituation over time. The first one was done on the SCR (the number of peaks above 0.05 microSiemens) observed after each stimulus, and the second one on the SCL, the tonic component of the skin conductance across the 24s trials. For SCR, after confirming normal distribution via the Shapiro-Wilk test, we performed a paired t-test to determine if the dynamic presentation led to a greater number of SCRs than the static presentation; and for SCL we computed a trial-triggered average by averaging skin conductance data across all dynamic and static trials for each participant, obtaining 24 s time-by-amplitude waveforms. Because skin conductance recorded over time does not follow a linear trend but rather follows a canonical function—characterized by a delay, a rising phase, and a decay phase—we used a Generalized Additive Model (GAM) to model these dynamics flexibly. Specifically, we used the “mgcv” package in RStudio (Wood, 2017), which implements GAMs with automatic smoothness estimation (Mixed GAM Computation Vehicle). This approach allowed us to capture nonlinear changes in skin conductance and assess differences between static and dynamic stimulus presentations over time.

The model was specified as:

where s(Time) represents a smooth function of the skin conductance, allowing for non-linear relationships, with separate smooths estimated for each stimulus presentation (static and dynamic). The model was fitted using Restricted Maximum Likelihood (REML) to optimize the smoothness parameters.

To assess the impact of anxious traits on SCL, we conducted Pearson correlation analyses examining the relationship between total STAI scores (the combination of scores on the STAI-Y1, or state, and STAI-Y2, or trait, questionnaires) and dynamic vs. static mean SCL in all participants. We removed one subject whose STAI score was more than 2 standard deviations above the mean (total STAI = 42).

Finally, we tested whether participants had better recognition of the images presented with a static or a dynamic presentation, by computing the hit rate, defined as the proportion of correctly identified “old” images (those presented during the task) among all old images. This was done separately for images presented in the two conditions. Using the Shapiro-Wilk test, we determined that the data for both conditions were not normally distributed (p < 0.05), and therefore used the Wilcoxon signed-rank test to compare recognition performance between the two conditions. Analyses were conducted in RStudio Version 2023.12.0.

Results

Demographics and Personality Questionnaires

Table 1 summarizes participant demographics. Questionnaires revealed a relatively normal range of stressors (STAI-Y1: mean = 13.43, SD = 3.27; STAI-Y2: mean = 13.57, SD = 3.39) and number of traumatic events (TLEQ: mean = 2.73, SD = 2.56) across subjects. When asked about exposure to stress, 11 participants reported being horror movie watchers, and 10 indicated that they would experience discomfort or distress if presented with either one or multiple of the following: sexual harassment and domestic violence (n = 6), shootings (n = 5), animal cruelty (n = 4), medically-related scenes (n = 3), violence (n = 2), and car crashes (n = 1). No participants expressed potential extreme discomfort with natural disasters, blood, or bug/snakes.

Table 1.

Demographics for the 16 enrolled participants

Measure
Age, mean (SD), years	27.31 (5.41)
Sex (Female), n (%)	8 (50)
Dominant hand (Right), n (%)	13 (81.25)
Racial background
Asian/Pacific Islander, n (%)	2 (12.5)
Black, n (%)	1 (6.25)
White, n (%)	13 (81.25)
Ethnicity
Hispanic, n (%)	2 (12.5)
Education
Bachelor’s degree, n (%)	10 (62.5)
Master’s degree, n (%)	6 (37.5)

Breath Holding

14 out of our 16 participants had increases in SCL when they held their breath for 3 s, thus classifying them as responders to the breath-holding task. One non-responder withdrew from the study before study completion. The other non-responder had neither a response to the breath holds, nor did they have any peaks in skin conductance for either condition in the main task. This subject’s data was excluded from the remaining analyses. An additional participant responded in the breath-hold task but their SCL did not show any variation during the main task. This participant was also excluded in the final analysis. This left us with a final sample size of 13 participants.

Skin Conductance Response

Given our small sample size, we conducted Shapiro-Wilk tests on the SCRs for the dynamic and static conditions and found that neither differed significantly from normality (Dynamic: W = 0.94, p = 0.43; Static: W = 0.87, p = 0.06). The resulting paired t-test showed that there was no significant difference in the number of SCRs between the dynamic and static conditions (Dynamic: mean = 60.62, SD = 35.89; Static: mean = 61.38, SD = 27.6; t = − 0.12, df = 12, p = 0.91).

Trial Triggered Average Over time

The generalized additive model (GAM) performed on the trial triggered average SCL explained a substantial proportion (49.1%) of the variance. Results revealed significant effects of the smoothed terms for Time, modeled separately for dynamic and static conditions (dynamic: edf = 8.954, F = 2692, p < 0.001; static: edf = 8.989, F = 3971, p < 0.001), indicating the expected non-linear variation of SCL over time. More importantly, the model also revealed a main effect of stimulus presentation type, indicating that SCL was significantly lower in the static than in the dynamic condition (β = − 0.003, SE = 0.001, t = − 4.53, p < 0.001). Indeed, as shown in Fig.3, while the group-averaged waveform of SCL in the static condition was initially higher than in the dynamic condition during the first four seconds of the trials (i.e., during the initial picture presentation), the decay of the SCL in the static condition was faster than for the dynamic condition. Further, while the SCL of the static condition stabilized after ~ 12 s, the SCL in the dynamic condition exhibited four additional small increases, which appeared to align with each new picture presentation.

Fig. 3.

Trial-triggered average of SCL for dynamic (orange) and static (blue) conditions. The shaded area represents the standard error

Impact of State and Trait Anxiety on SCL

A significant negative correlation was found between total STAI score and SCL in the static condition (r = − 0.62, p = 0.023), indicating that individuals with higher overall anxiety exhibited lower tonic autonomic arousal. In contrast, no significant association was observed in the dynamic condition (r = 0.40, p = 0.18; see Fig4). Given the small sample size and the high collinearity typically observed between state and trait STAI scores, we used the combined total score to reduce dimensional complexity and improve statistical stability. Analyses using separate state and trait subscales yielded no significant correlations in either condition.

Fig. 4.

Correlation of each participant’s combined STAI score with SCL in the dynamic (Orange) vs. static (Blue) presentation. Higher STAI score demonstrates a higher overall anxiety and is significantly negatively correlated with SCL only for the static condition (p = 0.023)

Recognition Memory Task

The Wilcoxon signed-rank test revealed a significant different in hit rate between the static (96 +/− 7%) and dynamic (78 +/− 14%) presentation methods (V = 120, p < 0.001). Participants better recognizted images presented in the static condition, which was expected given the longer time participants spent on looking at them.

Discussion

This study investigated whether dynamic image presentations, i.e., the displays of multiple aversive images in succession within a trial, would sustain autonomic arousal more effectively than static presentations of a single image, as measured by SCL. Given established associations between SCL and amygdala activation in fMRI research (Baczkowski et al., 2017; Cheng et al., 2006; LaBar et al., 1998; Phelps et al., 2001; Williams et al., 2001), we used SCL as a proxy for autonomic response and amygdala engagement. We hypothesize that if dynamic presentations limit habituation of the physiological fear response more than static presentations, dynamic presentations may be more desirable for usage in rt-fMRI-NF studies so as to limit confounding factors that could cause a misinterpretation of data. rt-fMRI-NF studies often rely on visual presentations for emotional arousal but have not systematically compared how stimulus presentation methods affect physiological arousal. Our findings indicate that presentation conditions significantly influenced physiological responses. Specifically, generalized additive modeling revealed that SCL was higher during dynamic presentations than static. As shown in Fig. 2, while the static condition elicited a sharper initial SCL over the first few seconds, the response decayed more rapidly than in the dynamic condition. In contrast, the dynamic condition showed more sustained SCL, with small increases following each new image onset.

These findings suggest that the dynamic format may prolong autonomic arousal by minimizing within-trial habituation. Interestingly, despite the greater initial increase in SCL in the static condition, the overall response across the full trial was stronger in the dynamic condition. This highlights two different patterns of habituation: within-trial (as the participant adapts to a stimulus over time) and between-trial (as the participant anticipates or adapts to upcoming stimuli). In dynamic trials, novel images continued to appear, minimizing the opportunity for adaptation. In contrast, static trials involved prolonged exposure to a single image, likely facilitating adaptation within each trial, but also potentially increasing sensitivity to the next novel image in the following trial. This interplay between within- and between-trial habituation underscores the complexity of temporal dynamics in psychophysiological studies and highlights the importance of careful experimental design.

Such nuances are particularly relevant to neuroimaging research. While the relatively high temporal resolution of SCL allows for more precise examinations of habituation, fMRI’s slower temporal resolution may obscure these patterns. This is especially critical for rt-fMRI-NF studies, where changes in physiological response might be misinterpreted as successful emotion regulation if stimulus habituation is not adequately controlled.

Our results contribute to the understanding of habituation to visual stimuli as we demonstrated that dynamic visual presentations sustained SCL more effectively than static ones, suggesting that habituation is an active and ongoing process in response to visual stimuli. This has important implications for studies relying on emotional picture viewing to elicit consistent autonomic responses, particularly in rt-fMRI-NF paradigms, where reduced physiological arousal could be mistakenly attributed to successful regulation rather than habituation to stimuli. Future research combining SCL with concurrent fMRI could clarify the neural correlates of this effect, particularly in regions such as the amygdala and central autonomic network. Such work would also help validate SCL as a high-temporal-resolution complement to BOLD signals, enriching the interpretation of emotion-related neuroimaging findings.

Image Recognition Task

In the image recognition task, participants recognized images presented in the static condition significantly more than those in the dynamic. This was expected, as participants viewed each static image for 24 s versus 4 s for each image in the dynamic condition. The recognition task acted as a verification that subjects were following instructions and attending to the aversive pictures presented. It also provided insight that dynamic presentation rates might allow for repeated image use while maintaining “novelty” and emotional provocation and limiting habituation. As such, it has implications for the design of neurofeedback studies, which usually consist of several sessions and are faced with the problem of a limited number of SEED images available.

Emotional Reactivity and SCL

Another important consideration in the design of experiments using aversive pictures is individual differences in emotional reactivity. In this study, we included the STAI to explore whether emotional reactivity correlates with SCL. Given the high collinearity between state and trait anxiety, and the fact that SCL is a measure of autonomic arousal influenced by both, we used the total STAI score to capture participants’ overall anxiety-related arousal. While we expected higher STAI scores to be associated with higher mean SCL, no correlation emerged in the dynamic presentation. However, we observed a negative correlation between STAI and SCL in the static presentation, suggesting that higher anxiety may be linked to reduced physiological reactivity to negative images.

Previous studies have reported mixed findings regarding the relationship between anxiety-related traits and skin conductance measures. For example, Mardaga et al. (2006) found that higher harm avoidance scores were associated with longer SCR half-recovery times for negative compared to neutral images. Positive associations have also been observed between skin conductance and anxiety-related traits, including state anxiety (Krusemark & Li, 2012). In contrast, other studies have reported no significant association between anxiety traits and skin conductance responses (Rosebrock et al., 2017; Martin-Soelch et al., 2006; Wood et al., 2014). To our knowledge, however, a negative correlation between overall anxiety scores and SCL, as observed in our static condition, has not previously been reported. One possible explanation is that the longer viewing time in the static condition may allow participants to engage in cognitive reappraisal strategies. Participants with higher STAI scores, though still subclinical according to their SCIDs, may be more experienced at managing their anxiety, resulting in reduced SCL. This finding should be interpreted with caution, given our limited sample size and the fact that we only observed this effect in the static condition. To better understand this relationship, future studies should include larger and more diverse samples, particularly to ensure a broader range of physiological responders, as many of our participants exhibited relatively low average SCL.

Limitations of this Study

Our study suggests that dynamic presentations limit overall habituation; however, we must acknowledge several limitations. First, as this was a pilot experiment to develop our experimental paradigm, our sample size was limited. Therefore, to confirm the conclusions from this work, a larger replication study is warranted. Additionally, we specifically recruited healthy adults with no history of psychiatric disorders, including anxiety and trauma-related disorders. It is possible that different patient groups may have differences in autonomic and amygdala responses to fearful stimuli; additionally, their habituation rates may differ. Therefore, future studies may want to examine habituation responses and validate this paradigm across clinical populations.

Second, while we initially expected that the dynamic presentation would elicit a greater number of SCRs than the static presentation, with roughly six times more SCRs in the dynamic condition, since each new picture was expected to evoke a response, we did not observe this difference. Instead, the number of SCRs was similar across both conditions. This lack of difference is probably due to our task design, which was optimized for neurofeedback study, and which did not include any interstimulus intervals (ISIs) between pictures, nor intertrial intervals (ITIs). It is possible that the skin conductance in the dynamic presentation did not have enough time to return to baseline before the next pictures was shown, therefore preventing observations of SCRs for the following pictures. Previous research has shown that skin conductance attenuates across trials unless a sufficiently long ITI (~ 20 s) is introduced (Benedek & Kaernbach, 2010; Breska et al., 2011). Interestingly, a recent review from Frumento et al. (2021) demonstrated that only studies using intertrial intervals greater than 20 s and stimulus duration greater than 20 ms successfully induced skin conductance reactions. While this review was based on subliminal presentation of aversive subliminal stimuli, it is possible that the same rationale applies to our study with longer stimulus presentation. The absence of an ITI in our design may have introduced between-trial carryover effects, potentially confounding our comparisons. A potential way to overcome this limitation could be to deconvolve overlapping SCR using a general linear convolution model as suggested by Bach et al. (2009).

Third, we combined aversive stimuli across a variety of categories, including snake/bug, animal, human, social, gore/body, and object/scene. This was done because of the limited number of fear-inducing/aversive images across SEEDs. However, it is likely that individuals vary in their emotional reactivity to different image categories. For example, Tan et al. (2013) found that skin conductance responses were stronger for pictures of living stimuli (e.g., animals) than non-living ones (e.g., guns), and responses were even more pronounced when the image involved a “human context.” Such individual differences in category sensitivity may have introduced uncontrolled variability into our SCL measures, complicating comparisons between presentation conditions. Therefore, there is a critical need for larger, updated SEEDs with which future studies could select stimulus sets tailored to participants’ specific emotional sensitivities or control stimulus categories more systematically.

Fourth, additional peripheral measures such as eye tracking and pupil diameter could help confirm participant engagement and offer complementary measures of emotional arousal. Incorporating such measures would strengthen the reliability of interpretations regarding attentional and emotional processes.

Conclusions & Future Directions

This is the first study, to our knowledge, to directly compare two rates of stimulus presentation to examine the effect of stimuli presentation method on skin conductance. Our findings suggest that dynamic presentations may better preserve reactivity by reducing within-trial habituation. Given prior work linking SCL to amygdala activation as measured by fMRI, our results underscore the importance of carefully considering stimulus presentation rate in fMRI studies of amygdala function. This is particularly relevant for rt-fMRI-NF studies, where a static presentation rates could inadvertently reduce amygdala activation, potentially mimicking volitional downregulation. Further, the use of SCL as a complementary measure in MRI studies may help clarify task dynamics. SCL offers a finer temporal resolution than fMRI and can aid in interpreting BOLD response, especially when paired with real-time imaging.

Author Contributions

HG – conceptualization, data curation, formal analysis, writing- original draft, writing-review& editing EW- data curation, formal analysis, writing- original draft, writing-review & editing LB – conceptualization, data curation, formal analysis, writing- original draft, writing-review& editing BL – conceptualization, supervision, review, and editing of final manuscript SHL -project administration, supervision, review, and editing of final manuscript. All authors approved the final version of the article. HG and EW share first authorship.

Funding

Gura H., Wiener E., Beynel, L., Luber, B., and Lisanby, S.H. are supported by the NIMH Intramural Research Program (ZIAMH002955). The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability

data are publicly available on open science framework: https://osf.io/nr7dw/.

Declarations

Conflict of interest

The authors declare no competing interests.