Abstract
Pre-stimulus alpha oscillations in the visual cortex modulate neuronal excitability, influencing sensory processing and decision-making. While this relationship has been demonstrated mostly in detection tasks with low visibility stimuli, interpretations of such effects can be ambiguous due to biases, making it difficult to clearly distinguish between perception-related and decision-related effects. In this study, we investigated how spontaneous fluctuations in pre-stimulus alpha power affect iconic memory, a high-capacity, ultra-short visual memory store. Data from 49 healthy adults (34 female and 15 male) was analyzed. We employed a partial report task, where a brief display of six stimuli was followed by a report cue indicating the target stimulus. In this paradigm, accuracy at short stimulus-cue onset asynchronies (SOAs) is typically high, reflecting the initial availability of sensory information, but it rapidly declines at intermediate SOAs due to the decay of the iconic memory trace, stabilizing at a low asymptote at long SOAs, representing the limited capacity of short-term memory. Crucially, performance in this task is constrained by the temporal persistence of sensory information, not by low visibility or response bias. We found that strong pre-stimulus alpha power enhanced performance by amplifying initial stimulus availability without affecting the speed of iconic decay. This effect partially reflects stronger pre-stimulus alpha power in the hemisphere ipsilateral to the to-be-reported target, likely suppressing neuronal excitability of neurons coding irrelevant stimuli. Our findings underscore the role of alpha oscillations in modulating neuronal excitability and visual perception, independent of decision-making strategies implicated in prior studies.
Significance Statement
Pre-stimulus alpha oscillations in the visual cortex are known to influence visual perception, but the exact mechanism has been debated. Our study reveals that spontaneous fluctuations in pre-stimulus alpha power, particularly alpha lateralization, enhance iconic memory—a brief, high-capacity visual memory system—by suppressing neuronal excitability at irrelevant spatial locations. This suppression improves the availability and temporal persistence of visual information and highlights a novel link between alpha oscillations and iconic memory. These findings extend our understanding of how pre-stimulus alpha power modulates neuronal excitability by showcasing its influence in a paradigm that is unaffected by low visibility and decision-making strategies.
Introduction
One of the most prominent neuronal signals in the human occipital cortex is the alpha rhythm, oscillating at around 8–13 Hz. Spontaneous pre-stimulus alpha power and phase fluctuations are associated with periodic fluctuations in neuronal excitability (Buzsaki and Draguhn, 2004; Haegens et al., 2011; Dougherty et al., 2017). These fluctuations affect visual perception, attention, and metacognition (Ergenoglu et al., 2004; Van Dijk et al., 2008; Limbach and Corballis, 2017; Samaha et al., 2017). Furthermore, these effects are often lateralized across the left and right cortical hemispheres. For instance, shifts of spatial attention due to external cues or internal decisions to attend are associated with alpha power increases in the hemisphere ipsilateral to the attended location, indicating relative inhibition of the hemisphere representing the unattended location, while alpha power decreases contralaterally, reflecting greater excitability in the hemisphere representing the attended location (Worden et al., 2000; Thut et al., 2006; Bengson et al., 2014). However, exactly how alpha power and excitability at the time of stimulus onset affect perceptual decision-making is still debated.
Studies testing the effect of pre-stimulus alpha power have demonstrated that weaker power is associated with higher hit rates in the detection of near-threshold stimuli (Ergenoglu et al., 2004; Van Dijk et al., 2008). Initially, this was interpreted as an improvement in the accuracy of visual detection; however, subsequent studies showed that the increase in hit rates accompanies an increase in false alarm rates, confidence, and subjective visibility, indicating a more liberal detection criterion (Benwell et al., 2017; Iemi et al., 2017; Samaha et al., 2017; Balestrieri and Busch, 2022). Interpreting these findings within the classical signal detection framework is challenging. A more liberal criterion can reflect a change in the observer’s deliberate decision-making strategy, independent of their subjective perception, or an amplified representation of both signal and noise, giving both a more signal-like subjective appearance (Samaha et al., 2020). Distinguishing between these interpretations is particularly difficult in the context of detection tasks with near-threshold stimuli, in which performance is limited by the discriminability of signal and noise, and observers apply a criterion for reporting the presence or absence of a stimulus. Here, we propose to overcome these problems by focusing on the effect of pre-stimulus alpha power on temporal stimulus availability rather than detectability.
Here, we investigated the previously unexplored effect of alpha oscillations on temporal persistence of visual information in iconic memory—a high-capacity, short-duration visual memory store that follows visual perception and precedes visual short-term memory (Sperling, 1960; Dick, 1974). We tested iconic memory performance and confidence using a partial report paradigm (Fig. 1A), in which a display comprising six concentric stimuli was briefly flashed, and the to-be-reported object was indicated by a cue that appeared after a variable stimulus-cue onset asynchrony (SOA; Lu et al., 2005). Performance in this task is excellent when the cue and display are presented simultaneously, implying that all the objects are available for report at cue onset. At longer display-cue SOAs, performance drops only gradually, implying that some stimulus information persists for a few hundred milliseconds after stimulus offset, reflecting the decay of information within a high-capacity iconic memory store (Gegenfurtner and Sperling, 1993). Teeuwen et al. (2021) demonstrated that iconic memory is based on the persistent firing of neurons in the primary visual cortex. Given that the strength of neuronal responses in the early visual cortex is correlated with the alpha rhythm (Dougherty et al., 2017; Lundqvist et al., 2020), we hypothesized that weak pre-stimulus alpha power (reflecting high neuronal excitability) would amplify the neural response or slow down its decay (Fig. 1B), thereby extending the persistence of information in iconic memory. Importantly, performance in this paradigm is not limited by sensory noise and is not influenced by a detection criterion. Thus, any effects of pre-stimulus alpha power on temporal persistence would reflect a genuine effect on perception, rather than strategic decision-making.
A, Illustration of the partial report paradigm. Each trial began with a central fixation cross displayed for a variable interval of 1,200–3,500 ms. Following fixation, a stimulus array consisting of six items was presented for 40 ms. Participants were then cued to report the orientation of one target item, with the cue appearing at various SOAs relative to stimulus onset: 140 ms before stimulus onset, at stimulus onset, or 40, 60, 80, 120, 200, 360, or 1,240 ms after. There was no time limit for reporting or the subsequent confidence rating. After the confidence rating, feedback was given via the color of the fixation cross, turning blue for a correct response and yellow for an incorrect one. B, Illustration of two possible mechanisms of temporal persistence modulation. Left: Schematic representation of the impulse response triggered by stimulus onset (indicated by the vertical gray bar). The response shows a rapid onset followed by a gradual decay. The stimulus information persists as long as the response amplitude remains above a critical threshold (shaded area). Horizontal bars indicate duration of persistence. Heightened excitability can extend this persistence either by increasing the amplitude of the initial response (top) or by slowing its decay (bottom). Right: Hypothetical performance in the partial report task as a function of SOA. Performance is high at short SOAs and declines with longer SOAs. An amplified response (top) is predicted to enhance initial stimulus availability, thereby boosting performance primarily at short SOAs, which would be captured by the model’s a1 parameter (Eq. 1). A slower decay (bottom) would improve performance at intermediate SOAs, reflected in the model’s τ parameter.
Material and Methods
Preregistration
The study was preregistered with the Open Science Framework (https://osf.io/u6h8d), and we adhered to the outlined methods unless stated otherwise.
Participants
We collected EEG and eye-tracking data from 61 healthy participants (aged 23.59 ± 3.94 years), of whom 45 were female and 16 were male. We collected more participants than stated in the preregistration due to favorable conditions. All participants had normal or corrected-to-normal vision, no reported history of neurological or psychiatric disorders, provided their written consent, and were compensated with either course credits or money. The ethics commission of the University of Münster approved the study (ref. 2023-27-PS). One participant was excluded due to poor EEG data quality resulting from a large number of uncorrectable EEG artifacts, and, five additional participants due to subpar performance (<80% accuracy at −140 and 0 ms SOA), and a further six participants because their performance data could not be fit with the exponential decay model (not preregistered; see below). The final sample size for the main analysis was 49 (aged 23.35 ± 4.1 years), with 34 female and 15 male participants, corresponding to the desired sample size stated in the preregistration. For the lateralization-based analyses, 8 more participants were excluded due to the experiment crashing or not completing it and thus not completing enough trials required for the modeling (<85%), leaving the sample size for this analysis at 41 (aged 23.34 ± 4.28 years): 27 female and 14 male.
Stimuli and procedure
The experiment was presented on a 24-inch Viewpixx/EEG LCD Monitor with a 120 Hz refresh rate, 1 ms pixel response time, 95% luminance uniformity, and 1,920 × 1,080 pixels resolution (33.76 × 19.38; www.vpixx.com). The recording took place in a dimly lit, soundproof cabin. Participants’ heads were stabilized on a chinrest with their eyes approximately 86 cm from the monitor. Eye movements were monitored using a desktop-mounted Eyelink 1,000+ infrared-based eye-tracking system (SR Research Ltd.) set to a 1,000 Hz sampling rate (monocular, from the participant’s dominant eye).
The stimuli used in the partial report paradigm were circles with wedges cut-out at variable orientations (0°, 45°, 90°, 135°, 180°, 225°, 270°, or 315°). The circles had a diameter of 2.6 degrees visual angle (dva) whilst the cut-out had a thickness of 0.28 dva with a diameter of 0.6 dva. The circles were light gray (RGB: [180 180 180]), and the cut-out wedges were of a darker gray (RGB: [70 70 70]), matching the background color. The cue utilized was a black line (RGB: [0 0 0]) with a length of 1 dva and a thickness of 0.1 dva, pointing to one of the former positions of the stimuli. Participants were instructed to avoid eye movements and blinks during the stimulus presentation. Participants had to fixate on the center of the screen, where a black (RGB: [0 0 0]) fixation cross with a size of 0.6 dva was presented for a variable time interval of 1,200–3,500 ms. Following this interval, six stimuli were arranged in a circle around the fixation cross and presented for 40 ms. The cue appeared after a variable SOA of either 140 ms before stimulus onset, at stimulus onset, or 40, 60, 80, 120, 200, 360, or 1,240 ms after stimulus onset. Participants were instructed to press a number on the number pad corresponding to the orientation of the wedge in the target circle (eight for 0°, six for 90°, two for 180°, etc.). Following this decision, the participants were instructed to press either four (low), five (medium), or six (high) on the number pad, indicating their respective confidence rating. There was no time limit for their response. Feedback on their decision was provided 300 ms after their confidence rating, with correct responses marked by a blue (RGB: [0 0 255]) fixation cross, and incorrect responses indicated by a yellow (RGB: [255 255 0]) fixation cross. A visualization of a trial can be seen in Figure 1. There were 1,296 trials, with short self-paced breaks after 170 consecutive trials to counterbalance SOA, stimulus position around the fixation cross, and target cut-out orientation. The experiment was written and presented using Matlab2022 (mathworks.com) and Psychtoolbox (Brainard, 1997; Pelli, 1997; Kleiner et al., 2007).
Modeling the time course of iconic memory
This study aimed to test the hypothesis that pre-stimulus alpha power and neuronal excitability modulate the temporal availability of stimulus information. We assumed that the response to a brief stimulus follows an impulse response, which can outlast the stimulus itself due to the low-pass filtering properties of the early visual system (Di Lollo, 1977; Loftus and Irwin, 1998). The longer the response amplitude remains above a task-specific critical threshold (Fig. 1B, shaded areas), the longer the stimulus information persists, remaining available for perceptual or decision-making processes. Greater neuronal excitability may enhance persistence in two ways: by amplifying the response amplitude and initial sensory availability (Fig. 1B, top), or by slowing the response’s decay (Fig. 1B, bottom). A similar concept has been proposed to explain the effects of oscillatory frequency on the temporal integration of double flashes (Karvat and Landau, 2024). To directly test the effect of alpha power on stimulus persistence and distinguish between its impact on response amplitude versus decay speed, we employed a partial report task. A display of six items was followed by a cue indicating the target to be reported (Fig. 1A). Intuitively, performance at short target-cue SOAs reflects the initial availability and response amplitude, while the performance decline across intermediate SOAs reveals the speed of iconic decay.
To quantify these effects and parameters formally, we analyzed the behavioral data using a signal detection theory framework, calculating d′ for each SOA. Behavioral data (d′ and confidence) was fitted to a nonlinear decay function:
EEG acquisition and preprocessing
EEG activity was recorded using a Biosemi Active Two EEG system with 67 Ag/AgCl electrodes (BioSemi B.V.) set to a 1,024 Hz sampling rate. Sixty-four electrodes were arranged in a custom-made montage with equidistant placement (EASYCAP GmbH) with additional external electrodes placed next to the right eye, the left eye, and below the left eye.
All EEG data preprocessing and analysis steps were scripted and run in Matlab2023a (mathworks.com) using the EEGLAB toolbox (Delorme and Makeig, 2004) and custom scripts. The continuous data was downsampled to 256 Hz and re-referenced to the average. It was then high-pass filtered at 0.1 Hz (preregistered was 0.5 Hz), and low-pass filtered at 40 Hz. Following this, the data was epoched from −1,000 to 1,500 ms time-locked to stimulus onset (preregistered was −400 to 200 ms). A round of trial rejection based on thresholding (±500 μV) and joint probability (function pop_jointprob in EEGLAB, local threshold nine, global threshold five) was performed. Furthermore, trials in which eyeblinks were detected in a range of −500 to 500 ms around stimulus onset, or in which participants significantly deviated from the fixation cross (2.5 dva) were rejected. An average of 32.653 trials were excluded per participant. After this, independent component analysis was applied, and components were identified by the IClabel algorithm (Pion-Tonachini et al., 2019) as “Brain,” “Muscle,” “Eye,” “Heart,” “Line Noise,” “Channel Noise,” or “Other.” Components were manually screened, and components classified as non-brain activity were excluded. Following this, noisy channels (SD > 2) were spherically interpolated. Three participants each had one channel interpolated.
EEG data analysis
Pre-stimulus spectral EEG power was computed using a Fast Fourier Transform (FFT) of the data within the pre-stimulus time range from −500 to −2 ms before stimulus onset for all electrodes and frequencies. Our analysis was focused on power averaged across electrodes corresponding approximately to P5, P3, P4, and P6 in the frequency range of 8–12 Hz. Single-trial power was then sorted into three bins using quantile ranking. Performance and confidence data were averaged across the trials in each bin for each non-negative SOA. The exponential decay model was then fit to d′ and confidence data separately for each bin. At this step, the start parameters of the decay function were chosen based on the best-fitting parameters of the global function fit described above.
In an exploratory analysis, we investigated the effect of pre-stimulus lateralization on performance. The rationale was that a spontaneous relative increase in alpha power in one hemisphere might impede the processing of stimuli represented in that hemisphere, i.e., of stimuli appearing in the contralateral hemifield. Such inhibition would reduce performance on trials where the upcoming cue required reporting a stimulus from the contralateral hemifield, and aid performance when an ipsilateral stimulus was cued. To this end, spectral power was quantified separately for trials with correct and incorrect responses to cued stimuli in the left and right hemifields based on the FFT analysis described above and on a time-frequency analysis using Morlet wavelets. A lateralization index was computed by subtracting contralateral pre-stimulus alpha power relative to the to-be-cued stimuli from ipsilateral pre-stimulus alpha power relative to the to-be-cued stimuli and normalizing this by dividing it through the sum of both (Thut et al., 2006):
Statistical analysis
The decay function parameters a0, a1, and τ were compared using t-tests between the weakest and strongest non-lateralized pre-stimulus alpha bins, as were the parameters of the weakest and strongest lateralization index bins, ipsilateral bins, and contralateral bins. In deviation from the preregistration, we decided against using a jackknife approach, as left-out single-trials make vastly different contributions to the different model parameters depending on the trials’ SOA.
To rule out any potential influence of the aperiodic activity of our signal on our analysis results, we decided to apply specparam (Donoghue et al., 2020). Note that our iconic decay model parameter estimates rely on single-trial binning, and specparam could not reliably model single-trial data or extract the aperiodic parameters for a single-trial. Therefore, we could not test the effect of the aperiodic parameters on our iconic decay model parameters directly. Thus, we extracted the aperiodic exponent and offset for the averaged correct and incorrect trials at our binning electrodes. We then used t-tests to compare the differences between these parameters.
Alpha power and performance have been shown to exhibit a time-on-task relationship, meaning that both measures can change systematically over the course of an experimental session, often due to factors such as fatigue, practice, or shifts in cognitive strategy (Benwell et al., 2019; Kopčanová et al., 2024). For instance, performance may improve as participants become more familiar with the task, while alpha power may increase due to changes in alertness or attentional state. To investigate potential time-on-task as a potential confound of our analysis, we ran a generalized linear mixed-effects model (GLME) across all trials and participants to predict single-trial pre-stimulus alpha power over the binning time window and at the electrodes used for the binning. The model included both accuracy (correct/incorrect) and time-on-task (trial order) as fixed effects, with subject ID as a random factor. The model specification was: Power ∼ Correct × TrialOrder + (1|Subject).
As alpha lateralization has been associated with eye movements and miniature gaze shifts (Van Ede et al., 2019; Popov et al., 2023; Mössing et al., 2024), we tested whether pre-stimulus alpha lateralization was confounded by pre-stimulus gaze shifts. Specifically, we tested whether correct responses to targets on the left or right side were preceded by pre-stimulus gaze displacements to the left or right of central fixation. To this end, we used a 2 × 2 repeated measures analysis of variance with cue direction (left/right) and correctness (correct/incorrect) as fixed factors, subject ID as a random factor, and gaze shift in dva relative to the fixation cross in the interval from 500 to 2 ms before stimulus onset as the dependent variable. In order to further scrutinize the potential effects of potential gaze shifts on performance, we investigated the spatial distribution of gaze shifts and their association with performance. We computed the two-dimensional histogram of gaze positions in the same time interval used for the analysis of alpha power (−500 to −2 ms). These histograms were then convolved with a 2D Gaussian (sigma = 15 pixels) to compute gaze density, represented as a heat map. We then tested for differences between gaze density for correct and error trials using two-sided t-tests.
Data and code accessibility
The data will be available upon acceptance of the manuscript at the Open Science Framework (osf.io/xfkrm/). The code will be available at github.com/pauljcs/alpha-iconic.
Results
Aperiodic activity
The t-tests revealed no significant differences in the aperiodic exponent between correct and incorrect trials [Channel P4: t(48) = 0.3158, p = 0.7529; Channel P6: t(48) = 0.1631, p = 0.8708; Channel P3: t(48) = 0.3244, p = 0.7463; Channel P5: t(48) = 0.4315, p = 0.6670] nor in the aperiodic offset between correct and incorrect trials [Channel P4: t(48) = 0.2328, p = 0.8164; Channel P6: t(48) = 0.2586, p = 0.7965; Channel P3: t(48) = 0.1842, p = 0.8543; Channel P5: t(48) = 0.1981, p = 0.8434].
Time-on-task
The GLME predicting single-trial pre-stimulus alpha power revealed significant main effects of both accuracy [F(1, 44, 992) = 9.22, p = 0.002] and time-on-task [F(1, 44, 992) = 10.42, p = 0.001].
Performance across all conditions
Bilateral pre-stimulus alpha power
Initial stimulus availability (a1) was higher for strong pre-stimulus alpha power than weak pre-stimulus alpha power [t(48) = −3.82, p < 0.001]. The function parameters representing the capacity for information transferred to short-term memory (a0) and the time constant representing the speed of iconic decay (τ) did not show significant differences between trials with weak versus strong pre-stimulus alpha power [a0 : t(48) = 0.53, p = 0.6; τ : t(48) = 0.16, p = 0.88; Fig. 2A].
Accuracy (d′) data for each SOA. Lines show model fits. Data are shown separately for A, trials with strong versus weak pre-stimulus alpha power; B, positive versus negative pre-stimulus lateralization, where positive lateralization implies lateralization “toward” the location of the upcoming target item; C, strong versus weak pre-stimulus power at ipsilateral channels, relative to the upcoming, cued target item; D, strong versus weak pre-stimulus power at contralateral channels. Model parameters showing a significant difference between these conditions are indicated with an asterisk.
For illustration purposes, we complemented this analysis by plotting the difference in power between correct and incorrect trials, irrespective of SOA. This analysis confirmed that trials with correct responses were preceded by stronger pre-stimulus alpha-band power (Fig. 4A).
Lateralized pre-stimulus alpha power
The parameter representing the initial stimulus availability (a1) was significantly higher for positive lateralization than negative lateralization [t(40) = −2.1176, p = 0.0403]. This indicates that performance was better if the ipsilateral to the to-be-cued stimuli pre-stimulus power was stronger relative to the contralateral to the to-be-cued stimuli pre-stimulus power. However, a0(the transfer to short-term memory) and τ (the speed of iconic decay) did not show significant differences [a0 : t(40) = 0.8316, p = 0.4105; τ : t(40) = 0.085, p = 0.9327] for positive and negative lateralization (Fig. 2B).
We complemented this analysis by plotting the difference between the correct versus incorrect effect for target items on the left from that for items on the right. This analysis confirmed a positive difference, reflecting stronger lateralization “toward” the upcoming cued item on correct trials (Fig. 5A).
To determine whether this effect is driven by (a) strong ipsilateral pre-stimulus power or (b) weak contralateral pre-stimulus power, we performed the same analysis separately for both ipsi- and contralateral pre-stimulus power.
Ipsilateral pre-stimulus alpha power
A significant difference was found for initial stimulus availability [a1; t(40) = −2.084, p = 0.0436] between weak and strong ipsilateral pre-stimulus alpha power, with strong ipsilateral pre-stimulus power resulting in better performance. However, no significant differences were found for the parameters representing the transfer to short-term memory (a0) or the speed of iconic decay [τ; a0 : t(40) = −1.073, p = 0.2898; τ : t(40) = 1.676, p = 0.1016; Fig. 2C].
Contralateral pre-stimulus alpha power
No significant differences were found for the initial stimulus availability (a1), transfer to short-term memory (a0), or speed of iconic decay [τ; a1 : t(40) = −1.574, p = 0.1234; a0 : t(40) = −1.192, p = 0.2402; τ : t(40) = 0.9702, p = 0.3378] between weak and strong contralateral pre-stimulus alpha power (Fig. 2D). Weaker contralateral pre-stimulus power appears not to affect iconic memory performance, whereas stronger ipsilateral pre-stimulus power does, thus indicating that the effect found for spontaneous lateralization does result from stronger ipsilateral pre-stimulus power.
Confidence across all conditions
Bilateral pre-stimulus alpha power
The function parameters representing initial stimulus availability (a1), information transfer to short-term memory (a0), and speed of iconic decay (τ) did not show any significant differences [a1 : t(48) = −0.7, p = 0.485; a0 : t(48) = 0.7, p = 0.487; τ : t(48) = −0.79, p = 0.434] for weak versus strong pre-stimulus alpha power (Fig. 3A).
Normalized confidence data for each SOA. Lines show model fits. Data are shown separately for A, trials with strong versus weak pre-stimulus alpha power; B, positive versus negative pre-stimulus lateralization, where positive lateralization implies lateralization “‘toward” the location of the upcoming target item; C, strong versus weak pre-stimulus power at ipsilateral channels, relative to the upcoming, cued target item; D, strong versus weak pre-stimulus power at contralateral channels. Model parameters showing a significant difference between these conditions are indicated with an asterisk.
For illustration purposes, we complemented this analysis by plotting the difference in power between high and low confidence trials, irrespective of SOA. This analysis confirmed that trials with high confidence responses were preceded by stronger pre-stimulus alpha-band power (Fig. 4B).
A, Topography and time-frequency plot showing the difference between correct and incorrect trials. Topographies represent the time-frequency window indicated by the black rectangle. Time-frequency plots represent the electrodes highlighted by the bold markers on the topography. B, Topography and time-frequency plot showing the difference between high and low confidence trials. Conventions as in A.
Lateralized pre-stimulus alpha power
No significant differences were found for initial stimulus availability (a1), transfer to short-term memory (a0), or speed of iconic decay [τ; a1 : t(40) = −1.0509, p = 0.2994; a0 : t(40) = 0.3436, p = 0.7329; τ : t(40) = 0.2707, p = 0.7880] for negative and positive lateralization (Fig. 3B). Again, we performed the same analysis separately for ipsi- and contralateral pre-stimulus alpha power in order to determine the cause of this effect.
We complemented this analysis by plotting the difference between the high versus low confidence effect for target items on the left from that for items on the right. This analysis indicated a weak, but positive difference, reflecting stronger lateralization “toward” the upcoming cued item on high confidence trials (Fig. 5B).
A, Topography shows the difference between correct and incorrect trials to target items on the left versus correct and incorrect trials to targets on the right within the time-frequency window indicated by the black rectangle. Time-frequency plot shows the difference in lateralization index between correct and incorrect trials at the electrodes highlighted by the bold markers on the topography. Warm colors represent stronger lateralization “toward” the target on correct trials, implying stronger power at ipsilateral versus contralateral electrodes. B, Topography shows the difference between high and low confidence trials to target items on the left versus high and low confidence trials to targets on the right within the time-frequency window indicated by the black rectangle. Time-frequency plot shows the difference in lateralization index between high and low confidence trials at the electrodes highlighted by the bold markers on the topography. Warm colors represent stronger lateralization “toward” the target on high confidence trials, implying stronger power at ipsilateral versus contralateral electrodes.
Ipsilateral pre-stimulus alpha power
No significant differences were observed for the function parameters representing initial stimulus availability (a1), information transfer to short-term memory (a0), and speed of iconic decay [τ; a1 : t(40) = 1.103, p = 0.2772; a0 : t(40) = −1.1015, p = 0.2778; τ : t(40) = −1.2143, p = 0.2318] for weak versus strong ipsilateral pre-stimulus alpha power (Fig. 3C).
Contralateral pre-stimulus alpha power
No significant differences were found for the function parameters representing initial stimulus availability (a1), information transfer to short-term memory (a0), and speed of iconic decay [τ; a1 : t(40) = 1.018, p = 0.3148; a0 : t(40) = −1.2168, p = 0.2308; τ : t(40) = 0.7259, p = 0.4721] between weak and strong contralateral pre-stimulus alpha power (Fig. 3D).
Gazeshift
The previous analyses showed that participants responded more accurately and with higher confidence when pre-stimulus alpha power was spontaneously lateralized “toward” the position of the upcoming item that would be cued for report. Given that alpha lateralization has been associated with miniature eye movements and gaze drifts (Van Ede et al., 2019; Popov et al., 2023; Mössing et al., 2024), we tested if small incidental horizontal gaze displacements before stimulus onset were systematically associated with better performance when the gaze shift was in the direction of the upcoming cued target item. Specifically, we tested if pre-stimulus gaze position was systematically displaced to the left on trials with a correct response to a left target, and vice versa. This interaction between correctness and cue direction was not significant, implying that accuracy was not confounded by gaze shifts [F(1) = 2.94, p = 0.0968]. The main effects of correctness [F(1) = 1.27, p = 0.269] and direction [F(1) = 0.21, p = 0.649] were also not significant. A descriptive inspection of the gaze density heatmaps shows no systematic difference in the spatial distribution of gaze density between correct and incorrect trials (Fig. S1). A statistical analysis using two-sided t-tests also revealed no statistically significant differences (Fig. S2).
Discussion
How does the ongoing state of neuronal excitability influence sensory information processing? We addressed this question by investigating the role of pre-stimulus alpha-band lateralization in modulating the temporal persistence of visual information in iconic memory. Specifically, we hypothesized that spontaneous alpha oscillations modulate the temporal persistence of sensory information by altering cortical excitability, thereby modulating iconic memory performance.
This was grounded in the theoretical framework suggesting that alpha oscillations reflect ongoing changes in neuronal excitability, influencing perceptual decision-making. Numerous studies (reviewed in Samaha et al., 2020) have demonstrated that near-threshold targets are less likely to be detected during periods of strong alpha power, reflecting reduced excitability (Ergenoglu et al., 2004; Van Dijk et al., 2008). Importantly, these effects are due to a more conservative detection criterion rather than impaired accuracy (Limbach and Corballis, 2016; Iemi et al., 2017). Thus, pre-stimulus alpha power has a much less pronounced effect in tasks not relying on detection criteria, e.g., feature discrimination tasks (Iemi et al., 2017). However, signal detection theory does not fully explain how changes in decision criteria arise, and the theoretical implications remain debated (Samaha et al., 2022). One interpretation is that pre-stimulus brain states reflect a decision bias: a post-perceptual adjustment in decision-making independent of subjective perception (Kloosterman et al., 2019). Another interpretation suggests a perceptual bias: a modulation of the intensity and subjective signal-likeness of sensory input and noise (Samaha et al., 2020; Benwell et al., 2022).
To address this ambiguity, we employed a criterion-free iconic memory task, where performance is constrained by temporal persistence of sensory information rather than near-threshold stimulus intensity. Iconic memory, a high-capacity storage system, temporarily holds visual information for a few hundred milliseconds before transferring some content to more durable, low-capacity short-term memory (Sperling, 1960; Neisser, 1967). It is supported by persistent firing in early visual areas (Duysens et al., 1985; Keysers et al., 2005; Teeuwen et al., 2021), where strong alpha power has been linked to reduced visual-evoked responses (Iemi et al., 2019; Lundqvist et al., 2020). Since the strength and duration of stimulus-evoked responses are critical determinants of iconic memory performance (Di Lollo, 1977, Fig. 1B), we predicted that strong alpha power, reflecting reduced excitability, would shorten stimulus availability and impair partial report performance.
Contrary to our hypothesis, stronger bilateral pre-stimulus alpha power was associated with better performance at short target-cue SOAs, suggesting amplified initial stimulus availability (Figs. 2A, 4A). No effects were observed on model parameters related to the speed of decay or transfer to short-term memory (Fig. 2A). Aperiodic parameters showed no difference between correct and incorrect trials, suggesting that this effect stems from oscillatory power differences. Confidence ratings were unaffected by pre-stimulus alpha power (Figs. 3A, 5B), contrasting with previous studies linking stronger alpha power to lower confidence in detection tasks (Benwell et al., 2017; Samaha et al., 2017).
To better understand this result, we conducted an exploratory analysis of pre-stimulus alpha lateralization, examining power separately in left and right channels and testing their effects on left and right targets independently. Stronger pre-stimulus alpha power specifically at electrodes ipsilateral to the upcoming target improved performance (Fig. 2C). For example, stronger alpha power in the left posterior channels enhanced performance for left but not right targets (Fig. 2B). In contrast, no lateralization effects were found at contralateral channels (Fig. 2D). The superposition of these two genuinely lateralized effects—strong left-lateralized power enhancing performance for left targets, and right-lateralized power enhancing performance for right targets—therefore contributes to the observed association between strong bilateral alpha power and improved performance (Fig. 6).
Topographical distribution of pre-stimulus alpha power effects. The channels used for the binning are marked in black. The color bar shows the amplitude difference in μV. A, Effect of accuracy (correct–incorrect responses) on trials with left cues. B, Effect of accuracy (correct–incorrect responses) on trials with right cues. C, Sum of effects shown in A and B. D, Effect of accuracy (correct–incorrect responses) without considering cued hemifield; this topography is identical to that in Figure 4A. Note that the sum of the lateralized effects shown in A and B strongly resembles the topography of the bilateral effect in D.
Could a time-on-task effect explain this association? Studies have demonstrated concurrent changes in behavioral accuracy and alpha power, often attributed to fatigue, practice, or shifts in cognitive strategy (Benwell et al., 2019; Kopčanová et al., 2024). Indeed, we find an additional effect of time-on-task on pre-stimulus alpha power. Whilst time-on-task may have influenced the effect of bilateral pre-stimulus alpha power on initial stimulus availability, the association between accuracy and power remained highly significant even after accounting for time-on-task, indicating that the observed relationship between stronger pre-stimulus alpha power and better performance cannot be fully explained by time-on-task alone. Additionally, time-on-task does not account for the lateralized effect: an alpha power increase over time would not differentially affect the hemispheres based on the hemifield of the upcoming target. In sum, while a time-on-task effect contributes to the relationship between accuracy and pre-stimulus alpha power, it neither explains nor accounts for effects of lateralized pre-stimulus alpha power on initial stimulus availability.
The effect of pre-stimulus ipsilateral alpha power on initial stimulus availability aligns well with research on cue-induced spatial attention, where alpha power increases over the hemisphere ipsilateral to a cue and decreases contralateral, reflecting greater inhibition in the hemisphere processing the irrelevant hemifield and heightened excitability in the hemisphere processing the relevant hemifield (Foxe and Snyder, 2011; Händel et al., 2011). In our study, pre-stimulus alpha lateralization was not externally driven by an attentional cue but occurred spontaneously, likely due to trial-by-trial fluctuations in self-initiated endogenous attention (Bengson et al., 2014; Nadra et al., 2023) or spontaneous neuronal excitability changes (Balestrieri and Busch, 2022).
Thus, the analysis of pre-stimulus lateralization supports the hypothesis that strong lateralized pre-stimulus alpha power enhances performance specifically for targets in the ipsilateral hemifield. This effect was linked to the model parameter representing initial stimulus availability, reflecting stimulus-evoked response strength, essential for iconic memory, visual persistence, and temporal integration (Di Lollo, 1977; Karvat and Landau, 2024; Fig. 1B, top panel). In turn, lateralization may reflect inhibition of the non-cued hemifield, reducing competition during stimulus encoding.
Together, these results demonstrate that spontaneous fluctuations in the momentary state of neurophysiological excitability and inhibition can modulate the temporal availability of sensory information. This supports previous models linking alpha oscillations to perceptual bias (Samaha et al., 2020) and suggests that detection of near-threshold targets may involve not only amplified stimulus-evoked responses but also prolonged availability due to more persistent responses.
Several avenues for future research emerge. First, our preregistered analysis of bilateral power produced the unexpected result that stronger alpha power was associated with improved performance. Our time-on-task analysis indicated that the time-on-task effect contributed to the observed bilateral effect to some extent. However, our exploratory follow-up analysis strongly suggests that this bilateral effect is better explained by the superposition of well-known lateralized effects (Fig. 6). This aligns well with the existing literature on cue-induced and spontaneous lateralization and their effect on cortical excitability—however, it also raises new questions. Many previous studies reporting that strong bilateral alpha power impairs detection have used lateralized target stimuli (Chaumon and Busch, 2014; Iemi et al., 2017; Pilipenko and Samaha, 2024). It remains unclear why these effects were not similarly lateralized in those studies. It is conceivable that the lateralization effect was more pronounced in the present study because it used a supra-threshold stimulus display and a task where performance was not limited by low stimulus visibility. However, only a direct comparison between detection and partial report tasks could help resolve this issue. Second, alpha power had no influence on confidence in our study compared to previous findings. This may be due to the high signal-to-noise ratio in our paradigm, where confidence is primarily driven by decision accuracy rather than uncertainty, as in near-threshold paradigms (Peters, 2022). Third, does the lateralized effect of alpha power on iconic memory performance stem from self-initiated spatial attention, and how does it compare to cue-induced attention effects? The relationship between attention and iconic memory has been debated: some authors argue that it requires little to no attention (Bachmann and Aru, 2016; Aru and Bachmann, 2017; Pinto et al., 2017), while others contend that it depends on attention (Mack et al., 2015, 2016), or is at least modulated by it (Botta et al., 2019). Comparing the effects of spontaneous lateralization on iconic memory to those of cue-induced lateralization would offer valuable insights not only into the role of the alpha rhythm but also into the broader theories of iconic memory, attention, and awareness.
Conclusion
Our study demonstrated that pre-stimulus alpha power influences iconic memory performance through spontaneous lateralization. Strong pre-stimulus alpha power ipsilateral to the to-be-cued stimuli suppresses neuronal excitability in visual areas representing task-irrelevant information, thereby extending the temporal availability of relevant information and improving accuracy. These findings suggest that alpha-induced modulations of excitability not only affect the detection of near-threshold stimuli but also enhance the persistence of visual information in iconic memory. This lateralized effect mirrors the role of cue-induced spatial attention in alpha modulation, contributing to the ongoing discussion on the relationship between attention and iconic memory. Ultimately, our work advances the understanding of how fluctuations in neuronal excitability shape both perception and sensory memory.
Footnotes
We thank Teresa Berther, Milena Carolin Koch, Seray-Ezgi Öztekin, Mathilde Maria Pöppelmann, Johanna Seroka, Simon Steibel, and Charlotte Wellmann for help with the data acquisition.
The authors declare no competing financial interests.
This paper contains supplemental material available at: https://doi.org/10.1523/JNEUROSCI.2117-24.2025
- Correspondence should be addressed to Paul Justin Connor Smith at paul.smith{at}uni-muenster.de.
