The phase coherence of cortical oscillations predicts dynamic changes in perceived visibility

Abstract

The phase synchronization of brain oscillations plays an important role in visual processing, perceptual awareness, and performance. Yet, the cortical mechanisms underlying modulatory effects of post-stimulus phase coherence and frequency-specific oscillations associated with different aspects of vision are still subject to debate. In this study, we aimed to identify the post-stimulus phase coherence of cortical oscillations associated with perceived visibility and contour discrimination. We analyzed electroencephalogram data from two masking experiments where target visibility was manipulated by the contrast ratio or polarity of the mask under various onset timing conditions (stimulus onset asynchronies, SOAs). The behavioral results indicated an SOA-dependent suppression of target visibility due to masking. The time-frequency analyses revealed significant modulations of phase coherence over occipital and parieto-occipital regions. We particularly identified modulations of phase coherence in the (i) 2–5 Hz frequency range, which may reflect feedforward-mediated contour detection and sustained visibility; and (ii) 10–25 Hz frequency range, which may be associated with suppressed visibility through inhibitory interactions between and within synchronized neural pathways. Taken together, our findings provide evidence that oscillatory phase alignments, not only in the pre-stimulus but also in the post-stimulus window, play a crucial role in shaping perceived visibility and dynamic vision.

cortical oscillations, EEG, inhibition, perceived visibility, phase

Introduction

The phase of brain oscillations reflects the state of ongoing neural activities and has been proposed as an essential factor for perception and cognition. Studies investigating the impact of the oscillatory phase on sensory processing span multiple scales. Previous research has shown that spiking, evoked, and even BOLD activities as responses to identical visual stimuli can vary depending on the phase of oscillations (Lakatos et al. 2008; Haegens et al. 2011; Scheeringa et al. 2011). Therefore, the phase of ongoing oscillations may establish a meaningful neural coding scheme (VanRullen et al. 2005; Fries et al. 2007; VanRullen et al. 2011), and a specific phase of neural oscillations can significantly influence perceptual and cognitive processes. Recent electroencephalogram (EEG) studies have demonstrated that the phase of ongoing oscillations can influence behavioral performance in various ways, including reaction times (Stefanics et al. 2010), saccadic latency (Drewes and VanRullen 2011), visual awareness (Busch et al. 2009; Mathewson et al. 2009), and attention (Busch and VanRullen 2010). Neural processing has been shown to be most effective when it occurs during specific phases of ongoing oscillations, while the opposite phases may suppress such activity (Mathewson et al. 2012; Merholz et al. 2022). Moreover, the phase of spontaneous oscillations can be locked or entrained by external factors, such as rhythmic visual stimuli (Mathewson et al. 2012) or non-invasive electrical stimulation (Thut et al. 2011).

A common approach to evaluate the role of phase relations of oscillatory activities in visual perception typically involves measuring the variability of spontaneous phases preceding stimulus onset or the phase alignment subsequent to stimulus presentation. Phase coherence emerges when the variability in phase distribution deviates significantly from uniformity within a group of trials. Numerous studies have highlighted that pre-stimulus phase coherence across different frequency bands can predict the perceptual outcome of the upcoming stimuli (Busch et al. 2009; Mathewson et al. 2009; Busch and VanRullen 2010; Benwell et al. 2017) and influence the likelihood of stimulus detection (Mathewson et al. 2011; Brüers and VanRullen 2017). While the role of pre-stimulus phase alignment in predicting the performance outcome of upcoming stimuli is well established, the modulatory effects of the post-stimulus phase on behavioral performance have been largely overlooked. Given this context, the current study aims to concentrate on the impact of post-stimulus phase coherence of cortical oscillations on perceptual task performance. As we delve deeper into the neural mechanisms underlying visual perception, it is essential to emphasize the role of oscillatory coherence in facilitating neuronal communication. It has been proposed that phase coherence between processing channels or neuronal groups facilitates effective communication through rhythmic and coordinated oscillations (Fries 2005, 2015). On the other hand, an alternative proposal argues that coherence may not be the cause but the result of communication (Schneider et al. 2021; Vinck et al. 2023). Both views, however, agree that oscillatory synchronization is an important neural marker, as the maximum excitability occurs during the cycles, resulting in temporal windows that favor communication. Previous research also suggests that these oscillatory cycles modulate the temporal segmentation of perceptual events and the timing of perceptual experience (VanRullen 2016a). Temporal windows, which are affected by the frequency, open and close rhythmically, thereby allowing for efficient interactions among neural groups. Investigation of the complex relationship between neural communication and coherence provides the foundation for their actual influence on perceptual tasks. Therefore, we argue that in order to understand the role of rhythmic synchronization in processing pathways of vision, it is essential to assess its influence on behavioral performance. More specifically, the phase coherence and its dynamics within (magno-dominated) transient and (parvo-dominated) sustained channels may provide novel insights into variability in visual discrimination abilities such that oscillatory networks in these processing pathways might synchronously route information from one area to another through coherent fluctuations.

Visual masking refers to the reduced visibility of one stimulus, called the target, due to the presence of a second spatiotemporally proximal stimulus, called the mask. A particular type of masking, called metacontrast, consists of the case where, temporally, the mask follows the target and, spatially, the mask does not overlap with the target. Because visual masking allows experimental control of target visibility, it has been extensively used to study brain correlates of conscious and unconscious vision. The dual-channel theory of masking suggests that inhibitory interactions occur between sustained and transient channels in response to the presence of a target and mask (Breitmeyer and Ogmen 2000, 2006). These interactions can inhibit sustained signals related to the stimulus visibility, depending on the strength of the mask and contour proximity. According to this approach, intra-channel (lateral and recurrent) inhibitions within the sustained channel also contribute to reduced stimulus visibility besides inter-channel inhibitory mechanisms (Ogmen 1993; Ogmen et al. 2003). Given these considerations, the modulatory effects of the presence of targets and masks on processing streams render metacontrast masking a useful paradigm for studying post-stimulus neural oscillations. More importantly, the simulations based on a dual-channel framework demonstrated that the recurrent connections within sustained channels can lead to oscillatory activities (Azizi et al. 1996; Purushothaman et al. 2000, 2003). These oscillatory activities with inter-channel inhibitions can also account for oscillations observed in masking functions at the perceptual level (Purushothaman et al. 2000; Fotowat et al. 2006). While there have been extensive neurophysiological recordings from the visual cortex and theoretical studies at the perceptual level, there is still a lack of comprehensive understanding of the relationship between oscillatory activities and perceived visibility.

In the current study, we aimed to shed light on the functional role of phase synchronization in perceived visibility and contour discrimination. We particularly focused on identifying the phase coherence of frequency bands that correlate with the changes in perceived visibility. We employed a metacontrast masking paradigm where target-induced post-stimulus phase alignments can be modulated by the presence of a mask to determine whether variations in phase coherence are associated with visual discrimination performance. Therefore, our primary objective was to elucidate how the proposed theoretical framework and interactions manifest in the phase coherence patterns of oscillations. We hypothesized that alterations in frequency-band-specific post-stimulus phase coherence serve as an indicator for the suppression in perceived visibility. To comprehensively test this hypothesis, we introduced variations in the contrast level of the mask into the experimental design to manipulate the masking-induced changes in perceived target visibility. In this way, we examined the relationship between the changes in phase coherence and perceived visibility. We performed further analyses to reveal modulations in the induced and evoked power of oscillations. To derive robust interpretations regarding the predictive role of phase coherence in altering perceptual outcomes, we additionally analyzed another EEG dataset from a previous study (Aydin et al. 2021). This metacontrast study also used a similar experimental design, but the polarity of the mask was manipulated rather than the contrast level.

Materials and methods

Behavioral pre-study

In metacontrast, the suppressive effects of the mask on the target depend on the onset timing between the target and mask (stimulus onset asynchronies, SOA), and this dependency forms a U-shaped masking function. Also, other stimulus parameters and criterion contents significantly alter optimum masking (SOA) conditions and the specific morphology of the masking function (Breitmeyer et al. 2006, 2008; Breitmeyer and Ogmen 2006; Bachmann and Francis 2014). For the main EEG experiment, we aimed to use a few SOA conditions that were critical for the masking function. Accordingly, we first evaluated masking functions across a full range of SOA values for each contrast level of the mask in a behavioral pre-study. We identified key SOA conditions from the behavioral results of this pre-study and used the same experimental design and criterion content in the main EEG experiment.

Participants

Nine healthy human volunteers (six females, age range 18–30 years, two left-handed) participated in the pre-study. None of the participants had a history of neurological disorders by self-report, and all had normal or corrected-to-normal visual acuity. Prior to the experimental session, each participant was informed about the procedures, filled out a pre-screening form, and signed an informed consent form. The inclusion and exclusion criteria were established before data collection. One participant was excluded from further analysis due to a failure in performing the discrimination task and having an accuracy score below the threshold level of 75% in the target-only (T) condition. All procedures were in accordance with the Declaration of Helsinki (World Medical Association, 2013) and approved by the local ethics committee at Bilkent University.

Apparatus and stimuli

We used MATLAB (The MathWorks, Natick, MA, USA) with the Psychtoolbox 3.0 (Brainard 1997; Pelli 1997; Kleiner et al. 2007) extension to generate visual stimuli and to control timing and data collection. The stimuli were presented on a 20-inch display CRT monitor (Mitsubishi Diamond Pro 2070sb, 1280 × 1024 pixel resolution with 100 Hz refresh rate). A chinrest was used to stabilize the head position at a viewing distance of 57 cm. A SpectroCAL photometer (Cambridge Research Systems, Rochester, Kent, UK) was used to calibrate the screen and acquire a gamma-corrected lookup table for a linearized digital representation of the stimuli. The physical timing and duration of visual stimuli were confirmed with a photometer connected to a digital oscilloscope (Rigol SD 10204B, GmbH, Puchheim, Germany). Behavioral responses were collected with a standard keyboard. The experiment room was dimly lit and silent.

Based on the findings by Thaler et al. (2013), a black fixation that combined a bull’s eye and cross-hair was used to minimize eye movements and enhance fixation stability. The concentric fixation circles were defined by inner and outer diameters measuring 0.2° and 0.6°, respectively. The centers of the target and mask were precisely aligned and vertically positioned 3° above the fixation (Fig. 1A). The target was a disk of 1.5° diameter and had a 0.15° wide left or right contour deletion. The mask was a surrounding ring with inner and outer diameters of 1.55° and 2.55°, resulting in a 0.05° separation between the target and mask. This target-mask (TM) separation was adjusted to prevent a merged appearance of target and mask, particularly at short SOAs, which might cause the contour deletion to pop out. After debriefing, all participants reported that the target and mask were perceived separately in all SOA conditions. The target had a fixed luminance of 30 cd/m² over a uniform background (15 cd/m²), leading to a light gray stimulation. The mask luminance was adjusted based on the mask-to-target contrast ratio (M/T ratio), which was set at either 0.5 (low) or 3.0 (high) (Breitmeyer et al. 2006). Using the Weber contrast (Breitmeyer et al. 2008), the mask had a luminance of either 22.5 cd/m² (M_Low) or 60.0 cd/m² (M_High). Both target and mask stimuli were brighter than the background and had an identical duration of 20 ms.

Fig. 1

Experiment design and behavioral results. (A) Schematic representations and the timeline of visual events in TM conditions of contrast ratio and polarity datasets. There was either left or right contour deletion on the target and participants performed a contour discrimination task in these conditions. The target disk was surrounded by the mask annulus, and both were positioned 3° vertically above the fixation point. (B) Results of behavioral pre-study (n = 8). The percent correct difference between mask and baseline (Δ performance) for each SOA and M/T contrast ratio condition. The T condition corresponds to the baseline zero level (dashed line). The masking magnitude (i.e. a decrease in perceived target visibility) was quantified with the Δ performance values and displayed as a function of SOA for low and high M/T contrast ratio conditions. Error bars represent the standard error (±SEM) across subjects. (C) Behavioral results of EEG experiments. The percent-correct difference between mask and baseline (Δ performance) is shown for the two experimental datasets. Top: Contrast polarity (same vs. opposite) and SOA conditions (n = 14). Bottom: Contrast ratio (low vs. high) and SOA conditions (n = 16).

Design and procedure

Previous studies on metacontrast (Breitmeyer et al. 2006; Aydin et al. 2021) demonstrated that a task on contour deletion efficiently captures the overall morphology of masking function and reflects important changes in perceived target-visibility (see also Tata 2002). Following the terminology used in the masking literature (Breitmeyer et al. 1974; Breitmeyer 1978; Breitmeyer et al. 2006), we referred to this task as “contour discrimination” to differentiate it from surface- or brightness-based judgments. Accordingly, we utilized a two-alternative forced-choice contour discrimination task, wherein observers reported the right or left contour deletion of the target. We employed a 2x9 repeated-measures design with two M/T contrast ratios (0.5 vs. 3.0; or equivalently, low vs. high CR) and nine SOAs (0, 10, 20, 40, 60, 80, 120, 160, and 200 ms). The design also included a baseline T condition (Fig. S1). Each experimental session consisted of 1,140 trials, with 1,080 TM conditions (60 trials per condition × 2 contrast ratios × 9 SOAs) and 60 T conditions. Participants completed all trials in six consecutive blocks on the same day. Each block contained a balanced mixture of these conditions, with a break midway to allow participants to rest their eyes. Participants were instructed to maintain fixation during the trials. All possible TM configurations, along with the task about the target, were introduced prior to the main experimental blocks.

As illustrated in Fig. 1A, each trial started with the presentation of fixation. After a variable time interval (500–1,000 ms), either a TM (TM_Low or TM_High) condition with a particular SOA or the T condition was shown. After the stimulus offset, the observers used a keyboard to report the contour deletion of the target (left vs. right). Participants were allowed to respond when the stimulus disappeared. The subsequent trial did not start until a response was acquired.

EEG study

Participants

Eighteen adult human volunteers who did not take part in the behavioral pre-study, participated in the main EEG experiment. Each participant completed a short practice session prior to the EEG experiment to become familiar with the set-up, stimuli, and task. Two participants were excluded due to excessive EEG artifacts caused by eye blinks (see EEG Recording and Preprocessing). The remaining sixteen subjects (11 females, age range 18–30, with a mean age of 25.5 years) were retained for further analysis. All other inclusion/exclusion criteria and ethical approval were the same as those used in the behavioral pre-study.

Design and procedure

According to the findings of our behavioral pre-study, three critical SOA values of 10, 80, and 200 ms were identified for the EEG experiment (see also Fig. 1B). Two contrast ratios (TM_Low or TM_High) were used for each SOA value, and hence the design included six TM conditions (2 contrast ratios × 3 SOAs). The design also included T, mask-only (M) (M_Low or M_High), and no-stimulus (NS) conditions (Fig. S1). These non-target conditions (M, NS) were included to particularly reveal nonlinear modulations in neural activities. In masking paradigms, observers typically perform a task on the primary target while the surrounding mask is passively observed (i.e. task-irrelevant/secondary stimulation). In other words, a significant masking effect demonstrates that the mask interferes and interacts with the visual processing primarily driven by the target. Therefore, most masking theories emphasize nonlinear interactions between the representations of target and mask (e.g. Ogmen 1993; Francis 2000, 2003). These additional conditions allowed us to identify significant deviations from the pure summations (i.e. violation of the additive model) and nonlinear interactions/modulations in the time-frequency power analyses of waveforms. The timeline of events during a trial is shown in Fig. 1A. As in the behavioral pre-study, each trial started with the presentation of fixation for a variable time interval (1,000 ± 150 ms). After the offset of visual events, the participants were allowed to respond within 1,000 ms in the target conditions. If no response was acquired within this time window, the trial was repeated later in the block. A variable (1,000 ± 150 ms) inter-trial interval was used. All other stimulus parameters were the same as those in the behavioral pre-study.

As the task demands varied between conditions, the sequential presentation of non-response (i.e. non-target) and response (i.e. target) trials could introduce potential confounds by requiring an additional decision about whether to respond in a trial. That is to say, when conditions are mixed, subjects have to determine first whether a trial contains a target and then decide to respond or not, which might lead to additional perceptual and cognitive processes. To avoid such potential confounds, all of the conditions were divided into target and non-target blocks, with the specific task being indicated at the beginning of each block. The target block included T and TM conditions, requiring participants to perform a two-alternative forced-choice contour discrimination task. In the non-target block, the additional M and NS conditions were presented. Since there was no target stimulus, the participants were only fixating and did not perform any contour discrimination task. The order of these blocks was randomized across participants. A total of 10 conditions were present in the experiment (6xTM, 1xT, 2xM, and 1xNS), and each condition was repeated 80 times. The order of conditions was randomized within a block, and each participant completed a session (i.e. two blocks) lasting ~1 h, with brief breaks.

EEG recording and Preprocessing

The EEG recording and preprocessing stages were similar to those described previously (Akyuz et al. 2020; Kaya and Kafaligonul 2021; Catak et al. 2022). A 64-channel MR-compatible EEG system (Brain Products, GmbH, Gilching, Germany) was used to record high-density neural activities. The placement of Ag/AgCl passive electrodes on the EEG cap was based on the extended 10/20 system. Two of the scalp electrodes, AFz and FCz, were the ground and reference electrodes, respectively. The electrode impedances were kept below 10 kΩ by applying a conductive paste (ABRALYT 2000 FMS, Herrsching–Breitbrunn, Germany) and monitored throughout an experimental session. Electrocardiogram (ECG) activity was also recorded by attaching an ECG electrode to the back of the subjects. The signals were acquired at a sampling rate of 1 kHz and band-pass filtered (0.016–250 Hz) online. The neurophysiological signals, event markers, and behavioral responses were stored with BrainVision Recorder Software (Brain Products, GmbH) for subsequent analysis.

The data were analyzed offline using BrainVision Analyzer 2.0 (Brain Products, GmbH) and our own custom MATLAB scripts (The MathWorks). The preprocessing of EEG data consisted of different stages. First, large and unused segments due to the breaks between blocks were removed, and the signals were down-sampled to 500 Hz. We identified bad or broken channels by examining the power spectrum density of each channel, and the identified channels were corrected using topographic spline interpolation (Perrin et al. 1989). To eliminate power line noise, the data were filtered through a zero-phase shift Butterworth band-pass filter (0.5–100 Hz, 24 dB/octave) and a 50-Hz notch filter (50 Hz ± 2.5 Hz, 16th order). Independent component analysis, utilizing the Infomax algorithm, was applied to remove common EEG artifacts such as eye blinks. The ECG electrode signal was used to eliminate the cardioballistic artifacts (Allen et al. 1998). Then, the data were segmented into trials (i.e. epochs) within 2 sec (−1,000 ms, 1,000 ms) windows centered at zero with stimulus onset (i.e. target-onset for T and TM conditions, mask-onset for M conditions and the corresponding time point in NS conditions). The trials were screened semi-automatically for undetected artifacts, such as epochs showing voltage changes less than 0.5 μV or more than 200 μV in 100 ms, and oscillations exceeding 50 μV/ms were marked as bad and rejected. After these standard preprocessing steps, on average, 1.58% of trials were rejected per condition. The percentage values for each condition are presented in Table S1.

Data analyses

Behavioral data analysis

The proportion of correct responses (i.e. performance values) was computed for all conditions of an observer. As in previous research (Breitmeyer et al. 2008; Aydin et al. 2021), the performance value of the T condition was subtracted from that of the TM condition for each contrast ratio and SOA to quantify masking-specific effects on perceived target visibility. A two-way repeated-measures ANOVA with contrast ratio and SOA as the main factors were applied to the difference (Δ) performance values.

Time-frequency analysis: phase coherence

In the experimental paradigm studied here, the participants performed a task on target visibility while passively being exposed to a subsequent mask. In other words, the target and mask were primary task-relevant and secondary task-irrelevant stimuli, respectively. This implies that the neural representation of the mask interacts and interferes with the primary processing of target visibility. This interaction may be achieved through changes in the phase of ongoing brain oscillations (e.g. phase-resetting). To quantify nonlinear modulations in cortical oscillations and identify the relationship between these modulations and changes in perceived visibility, we computed inter-trial coherence (ITC) for individual frequencies on a trial-by-trial basis (introduced as Phase Locking Factor in Tallon-Baudry et al. 1996).

For time-frequency decomposition, we used a wavelet transform in which the preprocessed EEG signal of each trial was convolved with complex Morlet wavelets, |${e}^{i2\pi tf}{e}^{-{t}^2/2{\sigma}^2}$|⁠, where t is time, f is the frequency from 1 to 70 Hz with 0.2 Hz frequency bins, and σ defines the width of tapering Gaussian in each frequency band, set according to 4/(2πf) (Cohen and Cavanagh 2011). This wavelet transform optimally balances the trade-off between time and frequency resolutions, providing higher temporal resolution at higher frequencies and higher frequency resolution at lower frequencies. On the other hand, for the lowest frequencies analyzed (e.g. < 10 Hz), the wavelet procedure substantially blurs the signal amplitude over time, which may result in low-frequency responses appearing to commence before stimulus onset (Chandran et al. 2016). The angle of the resulting complex signal at each time and frequency point would give the phase value. The phase coherence was calculated by vector averaging of the trial-based phases using the following formula (Busch et al. 2009):

$$ ITC\left(t,f\right)=\left|\frac{1}{k}{\sum}_{j=1}^k{e}^{i{\phi}_j\left(t,f\right)}\right| $$

In the formula, k represents the number of trials, and |${\phi}_j$| is the phase angle of the |$j$|th trial. Wavelet-transformed single-trial data (i.e. complex convolution coefficient) was normalized (at each time and frequency point) to an absolute value of 1 (i.e. unit vectors) before trial averaging. With this, each trial contributed equally to the subsequent average in terms of amplitude (Mercier et al. 2013). Perfect phase alignment across trials results in a coherence value of 1, and less consistent phases result in smaller values. As proposed by VanRullen (2016b), if the trial outcome is influenced by phase, the ITC for specific trial groups (e.g. conditions) must exceed the ITC of the null distribution. For a specific experimental condition, the ITC values were calculated for each electrode separately before being averaged within selected clusters of electrodes.

The significance of ITC was assessed using a streamlined procedure of permutation and z-score test (VanRullen 2016b), encompassing all time points from −600 ms to 600 ms and frequencies ranging from 1 to 70 Hz. In the first step, the data from all participants were pooled, regardless of condition, and then randomly assigned to one of two categories. This shuffling procedure was employed to construct the surrogate data and was repeated 1,000 times to obtain a null distribution. Then, under the null hypothesis, which posits no difference between specific conditions, the difference in averaged ITC values between two particular conditions was calculated for both real data and corresponding shuffled surrogates (VanRullen 2016b). The difference in ITC values between particular conditions was converted to a z-score by subtracting the mean of the null distribution from the real difference in ITC and then dividing it by the standard deviation of the null distribution (Merholz et al. 2022). This z-score was translated into a p-value using a cumulative distribution function, resulting in a time-frequency map of p-values. Multiple comparisons were corrected using cluster-based statistics, with an alpha (ɑ) threshold value of 0.01 applied to the p-values associated with each time-frequency point (Cohen 2014). False alarms were controlled at the map-level threshold using the supra-threshold cluster size test. After thresholding all statistical maps generated under the null hypothesis and identifying the largest clusters in each map, we obtained a distribution of the largest suprathreshold clusters expected under the null hypothesis. The final step of cluster-based correction involved thresholding the map of observed statistical values, identifying the clusters in the observed thresholded map, and removing any clusters below the 99% threshold (two-tailed) of the distribution of the largest clusters expected under the null hypothesis. A continuous significance (two-tailed P < 0.01) was indicated by black contours in the time-frequency plots.

Similar to behavioral results, our main analyses focused on identifying significant ITC changes in the time-frequency domain across TM conditions and relative to the baseline (T) conditions. We specifically aimed to reveal modulations of ITC that correspond to changes in discrimination performance. As detailed above, the main behavioral measure was the difference (Δ) performance values (see Behavioral Data Analysis). Therefore, additional correlation analyses were performed to examine the relationship between changes in mean ITC values within identified frequency-time windows over selected electrodes and the difference performance values of each participant. These analyses were performed using linear regression models, in which both the intercept and slope were the coefficients. By comparing conditions with cluster-based statistics and then performing correlation analyses on the difference values, we were able to circumvent possible/unforeseen confounding factors commonly existing in all the conditions and the temporal smoothing or blurring in the low-frequency range due to the wavelet procedure.

Time-frequency analysis: power

Building on previous findings (Naue et al. 2011; Cohen 2014), which suggest that modulations of oscillatory activity may be manifested in either high variability (an induced response) or being strictly phase-locked to stimulus onset (an evoked response), different calculation methods were required. The evoked power was analyzed by applying the wavelet transform to the difference ERP (event-related potentials). On the other hand, the total power was calculated by applying the wavelet transform to each trial first and subsequently averaging together the time-frequency power from all trials. This measure of total power comprises both the phase-locked and non-phase-locked parts of the response. The power was calculated by the squared absolute values of the complex numbers (i.e. complex convolution coefficient) obtained from the Morlet transform at each time and frequency point, as described above. As suggested by Cohen and Cavanagh (2011), to compare power across frequencies, time points, and subjects, the computed power was transformed (i.e. normalized) into a decibel (dB) scale by using the average power of pre-stimulus activity from −800 to −600 ms as a baseline. This interval was selected to better reflect any power modulations that may occur during the pre-stimulus period. The significance of time-frequency power representations was assessed using nonparametric cluster-corrected permutation tests with an alpha (ɑ) threshold value of 0.01 applied to the p-values associated with each time-frequency point, as described for phase coherence. Similarly, a continuous significance (two-tailed P < 0.01) was indicated by black contours in the time-frequency plots.

We first computed the total power for each TM, T, M, and NS condition separately. Next, we computed time-frequency representations of derived waveforms to isolate nonlinear neural interactions between the neural representations of target and mask and elucidate how these components are affected by contrast ratio and SOA (Del Cul et al. 2007; Fahrenfort et al. 2007; Aydin et al. 2021). Using this approach, we aimed to compare the power correlates of T with targets in TM sequences (TM_Low, TM_High). Nonlinear components of target activity were aimed to isolate from TM sequences by applying synthetic subtraction of corresponding M (M_Low, M_High) and T conditions. Thus, we achieved a synthetic summation of the target and mask (T + M) activity of corresponding conditions. Since M trials did not require motor responses, confounding factors due to the summation of common activities (e.g. complete elimination of motor response) were avoided. Previous studies also highlight that the relative onset timing of stimulation is crucial (Cecere et al. 2017; Kaya and Kafaligonul 2021). Therefore, during the synthetic summation process, the waveforms of the M condition were time-shifted to align with the mask onset in the corresponding TM sequences, according to the physical SOA. We applied an additional correction to the summed waveforms to minimize the influence of confounding factors in the power correlates. This was essential to prevent potential misinterpretation of interaction effects when contrasting these T + M with those of comparable TM. For instance, Walter et al. (1964) proposed that the pre-stimulus anticipatory slow potentials (i.e. contingent negative variations) might persist post-stimulus onset. Consequently, we subtracted the non-target stimulus (NS) activities from the synthetic summation of T and M trials to balance pre-stimulus common activity. The synthetic waveforms (T + M – NS) for each contrast ratio and SOA conditions were then computed.

Analysis on the additional dataset

To comprehensively evaluate the relationship between perceived visibility and phase coherence, we incorporated an additional EEG dataset from a previous metacontrast study, including varying mask contrast polarity conditions (Aydin et al. 2021). In this previous study, the fundamental properties of visual stimuli, experimental procedures, and data acquisition were almost the same as those described here. The shape and size of the fixation, target, and mask matched those used in our study. More importantly, both studies shared a key similarity in the duration of the target and mask stimuli, which were set at 20 ms. The only fundamental difference was the luminance values of the target and mask stimuli, which defined the contrast ratio and polarity. The luminance value of the background was fixed at 45 cd/m², leading to a uniform gray field. The target luminance was 80 cd/m², yielding white stimulation. To have the same and opposite polarity conditions, the mask had a luminance of either 80 cd/m² (white) or 10 cd/m² (black), respectively. For both polarity conditions, equal Weber contrasts were achieved with these luminance values (Breitmeyer et al. 2008).

The main EEG data consisted of fourteen healthy individuals (age range: 20–32 years), each completing a (2 × 3) repeated-measures design, in which the mask polarity (white/same: M_Same, black/opposite: M_Opposite) and SOA (10, 50, and 200 ms) were the factors. In addition to TM conditions (TM_Same or TM_Opposite for each SOA, see also Fig. 1A), the design also included T, M (M_Same or M_Opposite), and NS conditions. Similar to the contrast ratio study, these conditions were divided into target and non-target blocks. The conditions in each block and the block order were randomized. Each condition was repeated 60 times, resulting in a total of 600 trials for each participant across the experimental session (60 trials × 10 conditions). All other stimulus parameters and experimental procedures were consistent with those used in the contrast ratio experiment.

The EEG system and recording of signals were similar to those described above. We acquired the raw dataset and applied identical preprocessing steps and analyses of time-frequency decomposition to the EEG signals. Following these standard procedures, on average, 2.15% of trials were rejected per condition. The percentage values for each condition are presented in Table S2. Since Aydin et al. (2021) previously analyzed this dataset to uncover the ERP correlates of metacontrast masking with respect to contrast polarity, we limited our analysis specifically to reveal significant changes in phase coherence and power of neural oscillations.

Results

Behavioral pre-study

For each contrast ratio, the difference (Δ) performance values are displayed as a function of SOA (Fig. 1B). We observed a typical U-shaped (type B) masking function in both conditions. The difference values (i.e. performance values) dropped to a minimum of around 60–80 ms SOA, suggesting an optimum masking effect and the highest suppression in target visibility around these SOA values. For short SOAs (0, 10, and 20 ms), the performance and target visibility were high, suggesting an enhancement rather than a suppression. When the SOA was longer than 80 ms, the performance values increased, and the perceived visibility became closer to the level in the T (baseline) condition. The overall morphology of masking functions and optimum SOA values were consistent with those reported in previous studies (e.g. Breitmeyer et al. 2006). We performed a two-way repeated-measures ANOVA with 2 contrast ratios × 9 SOAs as factors on the difference performance values. Both the main effects of SOA (F_8,56 = 20.794, P < 0.001, |${\eta}_p^2$|= 0.748) and contrast ratio (F_1,7 = 25.502, P = 0.001, |${\eta}_p^2$| = 0.785) were significant. Moreover, the two-way interaction between SOA and contrast ratio was also significant (F_8,56 = 4.398, P < 0.001, |${\eta}_p^2$| = 0.386). Post-hoc pairwise comparisons at each SOA revealed significant differences (Bonferroni-corrected P < 0.05) between low and high contrast ratios at three SOAs: 40 ms (t₇ = 4.685, P = 0.002), 60 ms (t₇ = 3.844, P = 0.044) and 80 ms (t₇ = 4.685, P = 0.002). Among these SOAs, the largest suppression in the target visibility was at 80 ms for the high contrast ratio.

The overall effect of contrast ratio on masking function was in line with previous findings (Breitmeyer et al. 2006, 2008). Based on these behavioral results, we selected three critical SOA values of 10, 80, and 200 ms that represent the overall morphology of masking functions and the effects of contrast ratio on masking. For the smallest SOA, we selected 10 ms rather than 0 ms to restrict any spatiotemporal proximity and merging of target and mask. Moreover, Aydin et al. (2021) used 10 ms as the shortest SOA, and this selection allowed us to make a direct comparison across two studies.

EEG study: behavioral results

The trials excluded during the EEG preprocessing due to artifacts were removed from the behavioral data analysis. Consistent with pre-study findings, three SOA conditions captured the overall shape of the masking functions (Fig. 1C: bottom; see also Table S3 for raw percentage performance values). A two-way repeated-measures ANOVA indicated a significant main effect of SOA (F_2,30 = 28.886, P < 0.001, |${\eta}_p^2$| = 0.658). Importantly, the interaction between SOA and contrast ratio was also significant (F_2,30 = 6.317, P = 0.005, |${\eta}_p^2$|= 0.296). However, the ANOVA test did not reveal a significant main effect for the contrast ratio (F_1,15 = 0.779, P = 0.391, |${\eta}_p^2$| = 0.049). Post-hoc comparisons found a significant difference in target visibilities between low and high contrast ratios only at 80 ms SOA (t₁₅ = 3.208, P = 0.038, Bonferroni-corrected). The outcome of these statistical tests confirmed the modulations of perceived target visibility by SOA and contrast ratio during EEG recordings.

Similarly, Aydin et al. (2021) reported differential effects of SOA for each polarity condition (Fig. 1C: top, see also Table S4 for raw percentage performance values). For the same polarity condition, a typical U-shaped masking function was obtained, with target visibility (i.e. performance values) being most suppressed at intermediate SOAs of 50 ms. On the other hand, the largest suppression of target visibility in the opposite polarity condition was at the shortest SOA of 10 ms, and the perceived target visibility monotonically increased as the SOA between the target and mask became larger (type A masking function). The authors reported significant main effects of SOA (F_2,26 = 38.17, P < 0.001, |${\eta}_p^2$| = 0.746) and contrast polarities (F_1,13 = 111.9, P < 0.001,|${\eta}_p^2$| = 0.896). More importantly, there was a significant two-way interaction (F_2,26 = 77.78, P < 0.001, |${\eta}_p^2$| = 0.857). Follow-up pairwise comparisons revealed a significant difference between the two polarity conditions only at the 10 ms SOA (t₁₃ = −15.940, P < 0.001, Bonferroni-corrected).

Time-frequency decomposition: phase coherence

We computed the phase coherence (ITC) values from electrodes placed over occipital and parietal-occipital regions (including Oz, O1, O2, POz, PO3, PO4, PO7, and PO8) in both contrast ratio and contrast polarity studies. These regions were selected based on the known involvement of underlying cortical regions in conscious visual perception, as supported by previous literature (Romei et al. 2007, 2010; de Graaf et al. 2014; Brüers and VanRullen 2017; Aydin et al. 2021; Bauer et al. 2021). Additionally, we computed ITC values for all electrodes and generated topographical maps to visualize the distribution of phase coherence across the scalp. This comprehensive approach allowed us to examine both localized and global patterns of phase synchronization in relation to visual processing. In both datasets, we first calculated the raw inter-trial phase coherence for all conditions (Fig. S2). Then, we analyzed the ITC of the TM by statistically comparing them to the corresponding T condition (Figs. 2 and 3). We further compared the TM conditions across contrast ratio or polarity for each SOA. The outcome of these comparisons for the contrast ratio experiment is shown in Fig. 2. The cluster-based permutation tests identified two distinct frequency modulations associated with visual discrimination ability in the time-frequency domain. Notably, one of these modulations occurred around the 10–25 Hz (alpha-beta) range, while the other modulation was primarily in the 2–5 Hz (delta-theta) frequency range.

Fig. 2

Time-frequency plots of ITC, locked to the target onset for each experimental condition, including a pre-target period averaged across all subjects (n = 16). The plots display mean values over occipital and parieta-occipital electrodes. Dashed lines within the time-frequency plots indicate the onsets of the target and mask stimuli, respectively. Rows represent the particular SOA values of (A) 10 ms, (B) 80 ms, and (C) 200 ms. Columns represent the phase coherence difference between particular conditions: (left) TM at low contrast ratio and T (TM_Low vs. T), (middle) TM at high contrast ratio and T (TM_High vs. T), (right) TM at high and low contrast ratios (TM_Low vs. TM_High). Black contours enclose regions in which continuous clusters in the time-frequency domain were significantly different between specific conditions (two-tailed P < 0.01). Accompanying scalp maps depict the topographical distribution of these phase coherence differences within specified time and frequency windows of interest. The color bar for topographic maps was placed at the bottom right.

Fig. 3

Time-frequency plots of ITC, locked to the target onset for each experimental condition, including a pre-target period averaged across all subjects (n = 14). The plots display mean values over occipital and parieta-occipital electrodes. Dashed lines within the time-frequency plots indicate the onsets of the target and mask stimuli, respectively. Rows correspond to different SOA values of (A) 10 ms, (B) 50 ms, and (C) 200 ms. Columns represent the phase coherence difference between particular conditions: (left) TM at opposite contrast polarity and T (TM_Opposite vs. T), (middle) TM at same contrast polarity and T (TM_Same vs. T), (right) TM at opposite and same contrast polarities (TM_Opposite vs. TM_Same). Black contours enclose regions in which continuous clusters were significantly different between specific conditions (two-tailed P < 0.01). Accompanying scalp maps depict the topographical distribution of these phase coherence differences within specified time and frequency windows of interest. The color bar for topographic maps was placed at the bottom right.

In particular, the presence of the mask significantly increased the phase coherence for masked targets (TM) compared to unmasked ones (T) in the 15–25 Hz frequency range for both low and high contrast ratio modulations at 80 ms SOA (black contour, all P < 0.01). In this frequency range, the effect was most pronounced between 100–250 ms after target onset for the low contrast ratio (Fig. 2B, left) and between 200–400 ms for the high contrast ratio (Fig. 2B, middle). This notable increase in phase coherence for masked conditions with a high contrast ratio continued into the 400–600 ms time window but shifted to the 10–15 Hz frequency range. With a mask onset at 80 ms, the effect appeared immediately after the mask offset for the low contrast ratio. However, for the high contrast ratio, this effect was delayed by 100 ms and lasted longer. For both conditions, pre-stimulus phase coherence within this frequency range was not observed. We additionally found significant changes at the shortest SOA. When the SOA was 10 ms, there was a decrease in phase coherence values within the 10–17 Hz frequency range for both contrast ratios compared to the T conditions (black contour, all P < 0.01, Fig. 2A, left and middle). Within this frequency band, the disruption in phase coherence due to the presence of the mask intensified immediately after the mask offset and lasted until 400 ms. To better understand the association between phase coherence and masking amount, we also compared the TM conditions across different contrast ratios at each SOA. The behavioral results indicated a significant difference between two contrast ratio conditions only at 80 ms SOA. In line with behavioral results, the comparison of phase coherence values across contrast ratio conditions revealed only a significant difference for the 80 ms SOA (Fig. 2B, right; all P < 0.01). Specifically, the coherence was increased with the magnitude of mask contrast within the 10–15 Hz and the 250–400 ms interval post-target onset. Moreover, the lack of significant differences in phase coherence between TM conditions at other SOA values was consistent with observed changes in behavioral results.

We applied the same analysis approach to the contrast polarity dataset, and the outcome of cluster-based permutation tests is shown in Fig. 3. According to the behavioral results, the effects of masking were significant in both polarity conditions of 50 ms SOA and only in the opposite polarity condition of 10 ms SOA. The amount of masking significantly differed between the polarity conditions only when the SOA was 10 ms. For the ITC modulations at 50 ms SOA, we observed that the phase coherence was significantly increased for masked targets (TM) compared to unmasked ones (T) within the 10–20 Hz frequency range (black contour, all P < 0.01), similar to the contrast ratio experiment at 80 ms. In this frequency range, the increase in phase coherence started around 150 ms after the mask onset and persisted until 400 ms for the opposite polarity (Fig. 3B, left) and 600 ms for the same polarity conditions (Fig. 3B, middle). The peak values were observed around 18 Hz and 250 ms. On the other hand, there was a significant decrease in the same frequency range (10–25 Hz) when the mask polarity was opposite, and onset timing was 10 ms (Fig. 3A, left; all P < 0.01). Within this frequency band, the decrease in phase coherence due to the presence of mask became apparent right after the mask offset and was present up to 300 ms (cluster peak at 14 Hz and around 160 ms). We also compared the ITC values of the same and opposite contrast polarity conditions for each SOA. There were significant clusters in the time-frequency domain only for 10 ms SOA. The cluster centered around the 13–20 Hz frequency range that lasted from 100 ms to 400 ms. However, there was no significant difference in phase coherence for the other SOA conditions. Overall, the outcome of these comparisons was consistent with the observed effects of polarity on the behavioral performance values (i.e. masking amount) at each SOA (Fig. 1C, top).

To better understand and interpret the association between the phase coherence around the 10–25 Hz and performance values, we performed linear regression analyses, including both contrast ratio and contrast polarity datasets (Fig. 4). We selected time-frequency intervals to compute the average ITC differences (TM—T) for each participant based on the peak points within the significant contours. We used difference performance values as behavioral measures. We focused on the conditions in which there was a significant change in phase coherence values compared to those in the T conditions. Hence, we restricted these analyses to the 80 ms SOA of both contrast ratios, 50 ms SOA of both polarities, and 10 ms SOA of opposite polarity. In other words, at similar levels of suppression significantly elicited by mask, we assessed whether individual variation in perceived visibility was correlated with ITC modulations. According to previous studies pointing to different masking functions (type A vs. type B), the masking effects observed in the intermediate (i.e. 50–80 ms) and short (e.g. 10 ms) SOAs may be driven by different neural mechanisms. Indeed, this is illustrated by the dual-channel theory of masking (Breitmeyer and Ogmen 2000). According to this theory, inter-channel inhibition occurs between magnocellular and parvocellular pathways. On the other hand, the intra-channel inhibition within the pathways becomes dominant and drives the masking effects at short SOAs. Accordingly, we have separately analyzed the 10 ms SOA of the opposite polarity condition. The outcomes of regression analyses are displayed in Fig. 4. The analysis revealed a strong correlation between the modulations of phase coherence and behavioral performance values (R² adjusted = 0.151, P = 0.002; Fig. 4A) for the intermediate SOAs. As Δ phase coherence (i.e. ITC difference) decreased, the Δ performance values increased. The positive ITC difference reveals that TM has stronger phase coherence compared to T. The analyses also revealed a significant correlation in the 10 ms SOA of the opposite polarity condition (R² adjusted = 0.262, P = 0.043; Fig. 4B). The ITC difference became more negative with an increase in behavioral performance. The outcome of these analyses suggests that the changes in phase consistency in the 10–25 Hz frequency range are associated with the magnitude of suppression in target visibility, leading to a change in final behavioral performance.

Fig. 4

The scatter plots illustrate the relationship between mean phase coherence differences (TM—T) and Δ performance. In (A) and (B), each data point corresponds to individual subjects, and time-frequency intervals to compute the averaged ITC differences were selected based on the peak points in the 10–25 Hz (alpha-beta) frequency band of corresponding SOA values within significant contours. (A) a significant correlation (P = 0.002) at intermediate SOA values (high and low contrast ratio at 80 ms, same and opposite contrast polarity at 50 ms). (B) a significant correlation (P = 0.043) at 10 ms SOA of contrast polarity. In (C) and (D), averaged phase coherence values were calculated in the 2–5 Hz (delta-theta) frequency band and [−100, 500] ms time interval. Each data point corresponds to a unique condition characterized by a combination of contrast conditions and three SOA values. Vertical and horizontal error bars correspond to the variance across subjects (±SEM). Significant correlations for (C) contrast ratio (P = 0.001) and (D) contrast polarity (P = 0.012) datasets. In all the plots, the black solid line indicates the best linear fit and dashed curves denote the 95% confidence intervals on the linear fit. The distribution of averaged values with individual samples is provided in Fig. S3.

The presence of the mask also led to significant modulations in the delta-theta range (2–5 Hz). When the masking effect was low and target visibility was high, there was a significant increase in ITC values within this low-frequency window compared to the T conditions. For instance, there was a strong increase in both contrast ratio conditions at the SOAs of 10 ms and 200 ms (Fig. 2A and C, left and middle; black contour, all P < 0.01). The contrast polarity dataset indicated similar modulations. The enhancements in ITC values were present around 2–5 Hz for the same polarity at 10 ms and for both polarities at 200 ms SOA (Fig. 3A, middle; Fig. 3C left and middle; black contour, all P < 0.01). This increase in phase coherence began ~100 ms before the target onset and sustained throughout the trial. On the other hand, these enhancements were absent when the effects of masking became larger and target visibility was suppressed. There was even a marked decrease in ITC at this low-frequency range for intermediate SOAs in both datasets (contrast ratio: 80 ms, contrast polarity: 50 ms; black contour, all P < 0.01). To illustrate and objectively evaluate these modulations across conditions, we performed additional linear regression analyses on contrast ratio and polarity datasets. We specifically tested whether changes in behavioral masking functions (SOA dependency) were correlated with SOA-dependent changes in ITC values. We determined the mean ITC difference across participants within a 2–5 Hz frequency band spanning a [−100, 500] ms time window. The analysis revealed a robust correlation between the difference phase coherence (TM—T) and performance values for both contrast ratio (R² adjusted = 0.925, P = 0.001; Fig. 4C) and polarity conditions (R² adjusted = 0.779, P = 0.012; Fig. 4D). Notably, unmasked conditions exhibited greater phase coherence than their masked counterparts in 2–5 Hz. Our results underscore the pivotal role of increased phase coherence in this frequency band, suggesting its potential as a significant determinant in enhancing target visibility and contour discrimination, thereby improving behavioral performance.

Our analysis of ITC revealed significant clusters in the alpha-beta frequency band around 200 ms post-target-onset, while for the delta-theta band, significant clusters emerged even earlier, aligning with the target onset. These time ranges were well before the average reaction times (RTs) estimated, even according to the mask offset in all the TM conditions (Tables S3 and S4). The timing of identified clusters indicates that the neural processes associated with perception were captured well before the behavioral responses were recorded. The emergence of significant ITC patterns prior to the average RTs across all conditions strongly suggests that these neural activities reflect early perceptual processes, providing further evidence that the topographical distributions observed in our analyses are indeed indicative of perception-related neural dynamics.

In addition to the statistical comparisons across conditions, we also examined the raw ITC values of both datasets (Fig. S2). These results revealed robust phase coherence in the 2–20 Hz frequency range, particularly in the occipital and parieto-occipital regions, consistent with the observed patterns in TM and T comparisons. This strong coherence in raw ITC supports the validity of our findings, ensuring that the significant differences detected between conditions are reflective of genuine neural activity rather than noise. While the raw ITC data provide valuable context by confirming the presence of consistent oscillatory activity, the main focus of our analysis remains on the statistical comparisons, which more clearly delineate the specific effects of our experimental manipulations. Moreover, we compared ITC values of M conditions by applying cluster-based permutation tests for both datasets (contrast ratio: M_High vs. M_Low, polarity: M_Opposite vs. M_Same). The analysis did not reveal any significant clusters associated with the contrast ratio or polarity effects (Fig. S8, left), suggesting that the observed effects in the 2–5 Hz and 10–25 Hz frequency ranges are not driven solely by contrast or polarity changes in the M conditions. This highlights the nonlinear neural activities driven by the interaction of target and mask (TM).

Time-frequency decomposition: power modulations

To align with the ITC analysis and facilitate a more direct comparison of the results, we first performed a power analysis by comparing values of the TM conditions with those of the corresponding T conditions. The results of these analyses are presented in Figs. 5 and 6. There were significant power differences (all P < 0.01) in the 10 ms SOA of both the contrast ratio and polarity experiments (Figs. 5A and6A), as well as in the opposite polarity condition when the SOA was 200 ms (Fig. 6C). These power differences were consistent with the ITC findings, particularly in the alpha-beta frequency bands during the 100–300 ms time window. Although a power difference was identified in the delta-theta range for the 200 ms SOA and opposite mask polarity condition, this effect was not consistent across other conditions, suggesting its limited reliability. These observations align with the idea that measurable differences in visual-evoked fields can result from either changes in amplitude or variations in phase consistency (Wutz et al. 2014). Our findings indicate that the observed power differences are likely to be attributable to stronger phase synchronization rather than amplitude changes. This interpretation is further supported by the concomitant significant differences observed in both power and ITC time-frequency maps, indicating that stronger phase synchronization to target onset contributed to the observed effects. Moreover, we also observed time-frequency ranges where significant ITC differences did not correspond to any significant power changes, particularly in the alpha-beta bands within 150–350 ms for mid-range SOAs and in the delta-theta bands within [−100, 500] ms for short and long SOAs (as shown in Figs. 2 and 3). These findings highlight the crucial role of phase coherence in visibility suppression, which appears to operate independently of power fluctuations.

Fig. 5

Time-frequency plots of baseline-corrected power (dB), locked to the target onset for each experimental condition, including a pre-target period averaged across all subjects (n = 16). The plots display mean values over occipital and parieto-occipital electrodes. Dashed lines within the time-frequency plots indicate the onsets of the target and mask stimuli, respectively. Rows represent the particular SOA values of (A) 10 ms, (B) 80 ms, and (C) 200 ms. Columns represent the power difference between specific conditions: (left) TM at low contrast ratio and T (TM_Low vs. T), (middle) TM at high contrast ratio and T (TM_High vs. T), (right) TM at high and low contrast ratios (TM_Low vs. TM_High). Black contours enclose regions in which continuous clusters in the time-frequency domain were significantly different between specific conditions (two-tailed P < 0.01). Accompanying scalp maps depict the topographical distribution of these power differences within specified time and frequency windows of interest. The color bar for topographic maps was placed at the bottom right.

Fig. 6

Time-frequency plots of baseline-corrected power (dB), locked to the target onset for each experimental condition, including a pre-target period averaged across all subjects (n = 14). The plots display mean values over occipital and parieto-occipital electrodes. Dashed lines within the time-frequency plots indicate the onsets of the target and mask stimuli, respectively. Rows correspond to different SOA values of (A) 10 ms, (B) 50 ms, and (C) 200 ms. Columns represent the power difference between specific conditions: (left) TM at opposite contrast polarity and T (TM_Opposite vs. T), (middle) TM at same contrast polarity and T (TM_Same vs. T), (right) TM at opposite and same contrast polarities (TM_Opposite vs. TM_Same). Black contours enclose regions in which continuous clusters were significantly different between specific conditions (two-tailed P < 0.01). Accompanying scalp maps depict the topographical distribution of these power differences within specified time and frequency windows of interest. The color bar for topographic maps was placed at the bottom right.

Following these initial analyses, we computed the total power for the TM and (T + M – NS) across each contrast ratio, polarity, and SOA value. This analysis was performed using electrodes located in the occipital and parieto-occipital regions, consistent with those used previously. Although there were time-varying frequency band–specific power modulations, we found no significant differences between TM and (T + M – NS) in any comparisons across the conditions of contrast ratio or contrast polarity experiments (Figs. S4 and S5, left and middle panels). Moreover, since our behavioral results revealed a significant difference in masking magnitude between the low and high contrast ratios at 80 ms, and same and opposite contrast polarities at 10 ms, we further tested the significance of power differences between those conditions. However, there was no significant difference in total power between those conditions at any of these SOAs (Figs. S4 and S5, right panel).

Similarly, we examined evoked power modulations on the derived ERP waveforms of TM and (T + M – NS) to focus more on the phase-locked power modulations (Figs. S6 and S7). We first tested the significance of evoked power differences between the masked target (TM) by comparing it to the corresponding synthetic summation (T + M – NS). Then, we further compared the derived difference waveforms [i.e. TM − (T + M − NS)] across two contrast ratios and polarities for each SOA. Statistical tests revealed no significant difference in evoked power between those conditions at any of these SOAs in contrast ratio and polarity experiments.

We also assessed evoked and total power modulations of M conditions for both datasets (contrast ratio: M_High vs. M_Low, polarity: M_Opposite vs. M_Same). Cluster-based permutation tests did not reveal any significant cluster associated with the contrast ratio or polarity effects (Fig. S8, middle and right). In addition to the statistical comparisons between the aforementioned conditions, we probed the power dynamics within each condition separately (Fig. S9). These analyses revealed that the visual stimulation led to notable alterations in the power of neural activities, particularly in the 2–15 Hz range, even though no specific power effects were statistically associated with the suppressed visibility. This suggests that while our comparisons did not yield statistically significant differences between conditions, the visual stimulation had a measurable impact on power dynamics. These power results provide valuable insights into the overall patterns of neural activities and complement our main analysis by confirming the presence and strength of the underlying oscillatory activity. However, they do not directly delineate the differences between conditions.

Discussion

Brain rhythms serve as important instruments for sensory and perceptual processing, with ample evidence supporting the functional role of the oscillatory phase in neural communication and behavior (Engel et al. 2001; Buzsáki and Draguhn 2004; Sauseng and Klimesch 2008; Thut et al. 2012). In the present study, we investigated the relationship between post-stimulus phase coherence and perceived visibility. The analyses on two different datasets revealed a significant correlation between the frequency-specific oscillatory phase coherence over occipital/parieto-occipital regions and changes in discrimination performance associated with perceived target visibility under different masking conditions. Notably, we found that the phase coherence of 10–25 Hz in the post-stimulus window was associated with the magnitude of masking in the intermediate SOA values. Furthermore, our results indicated significant modulations in the 2–5 Hz band, where phase coherence increased for the SOA conditions linked with high performance values and target visibility. Despite prior studies suggesting a role for pre-stimulus oscillations in predicting human perceptual performance (e.g. Romei et al. 2008; van Dijk et al. 2008; Di Gregorio et al. 2022), we found no evidence for the pre-stimulus phase coherence predicting the upcoming discrimination performance. Also, our analyses did not reveal any significant power modulations correlated with changes in performance and perceived visibility. The consistent patterns observed across two different metacontrast datasets provide supporting evidence for the essential role of frequency-specific phase dynamics in shaping perception. These findings have important implications for both theoretical and empirical investigations on visual masking and sensory processing.

Oscillatory phase and inhibitory mechanisms of vision

In the dual-channel model of masking, input is processed by the magno-driven transient and the parvo-driven sustained channels (Breitmeyer 1984; Ogmen et al. 2003). The perceived visibility and contour processing of an object is associated with the sustained activities at the cortical level (Breitmeyer et al. 2006). The model postulates that a reciprocal inhibition exists between sustained and transient channels. This reciprocal inhibition is referred to as inter-channel inhibition to distinguish it from inhibitory interactions (lateral and recurrent) within each channel, which in turn is called intra-channel inhibition. These two inhibitory mechanisms play a major role in perceived visibility and in determining the shape of masking functions. The model illustrates that the transient and sustained activities of the mask suppress the sustained (visibility) activity of the target by inter- and intra-channel inhibition, respectively. Although two inhibitory mechanisms are effective in metacontrast, the inter-channel inhibition becomes dominant, particularly in the intermediate SOA values (e.g. 50–80 ms range). When the mask was introduced with these SOA values, the mask led to a phase coherence increase in the 10–25 Hz range. The magnitude of this increase (compared to the T condition) was correlated with the magnitude of masking (i.e. a decrease in performance values). Therefore, it can be hypothesized that these changes in phase coherence may be mainly associated with inter-channel inhibitions. Interestingly, the mask led to an opposite change and a decrease in the phase coherence of this alpha-beta frequency range when the SOA was 10 ms and the mask had an opposite contrast polarity. This decrease in phase coherence (compared to the T condition) was also correlated with the decrease in performance values. In the opposite polarity condition, a monotonic type A function was observed (Fig. 1C). This function mainly highlights sustained parvocellular activity and the intra-channel inhibition at short SOAs. That is to say, the monotonic type A function has been associated with the stronger sustained signals elicited by the mask and the interference of this activity with that of the target. Therefore, one can hypothesize that a decrease in the phase coherence may mainly reflect intra-channel inhibition at short SOA values. Our findings suggest that distinct modulations of phase coherence in the 10–15 Hz range may indicate different inhibitory mechanisms underlying metacontrast masking. Furthermore, we observed SOA-dependent modulations of phase coherence in the delta-theta frequency range. When the perceived visibility of the target was high at short and long SOAs, there was a significant increase in the phase coherence of 2–5 Hz compared to the T condition. On the other hand, there was even some decrease around 5 Hz when the perceived visibility was low (e.g. 50–80 ms of SOAs). Such distinct modulations over various frequencies may reflect different mechanisms involved in perceived visibility and in the dynamics of sensory and perceptual processing.

Alpha-beta phase and temporal dynamics of vision

Previous studies showed that sensory processing and awareness vary with respect to the phase of the ongoing alpha rhythm (Lakatos et al. 2008; Busch et al. 2009; Mathewson et al. 2009). Of particular relevance to the current study, the perceived visibility of a masked target in metacontrast fluctuates as a function of alpha power and phase (Mathewson et al. 2009, 2011, 2012). With a metacontrast masking paradigm, Mathewson et al. (2009) demonstrated that oscillations over posterior brain regions reflect differences in visual target detection and awareness, with alpha oscillations exerting an inhibitory influence on target detection in a non-uniform manner. The findings also indicated an important relationship between the pre-stimulus alpha phase and perceived visibility. Using metacontrast with a real-time phase-locked stimulus presentation approach, Tseng et al. (2023) have recently investigated the relationship between the real-time phase of ongoing oscillations and behavioral performance. This approach was based on online presentations of masked target that were phase-locked to different stages of the alpha cycle while observers performed a detection task. Even though the findings failed to reveal an overall effect of the peri-stimulus alpha phase on target detection, further analyses indicated that the phase-related changes in perception depend on an early interaction between the phase and incoming stimuli. In the present study, we extend previous findings by pointing to a significant and intricate relationship between post-stimulus phase dynamics and perceived visibility. Our results particularly provide important evidence that changes in the alpha-beta phase are associated with masking magnitude, suggesting that the phase of alpha-beta bands may reflect inhibitory mechanisms of masking.

Previous studies also reported that the alpha phase takes place in the temporal integration and segregation of brief visual events. Of interest, Wutz et al. (2014) used integration masking in a forward/simultaneous masking paradigm to investigate a possible link between phase synchronization and temporal integration windows. The experimental design was based on comparing trials in which observers accurately segregated target items from mask persistence with those in which target elements were integrated in time. Correct segregation was observed when the SOA was longer than 100 ms and accompanied by a phase synchronization (i.e. phase reset elicited by the mask) in the alpha band. Several M/EEG studies using different paradigms confirmed that the phase of ongoing oscillations in the alpha band predicts whether successive stimuli are integrated or segregated in time (Milton and Pleydell-Pearce 2016; Wutz et al. 2016, 2018; Ronconi et al. 2017, 2018). Although there is strong evidence that the alpha phase plays an important role in temporal segregation/integration, some of these studies additionally indicate the involvement of pre-stimulus beta power and theta (7–8 Hz) phase in the temporal organization of brief visual events (Wutz et al. 2014; Ronconi et al. 2017; Ronconi and Melcher 2017). A recent review on beta oscillations has highlighted that tasks requiring spatial segregation/integration, such as contour integration, and the organization of visual inputs rely on beta oscillations, primarily originating from parietal areas (Di Dona and Ronconi 2023). Specifically focusing on low-beta (~15–18 Hz) rhythms generated mostly by lateral occipito-parietal regions, Di Dona and Ronconi (2023) suggested that beta rhythm is associated with both the magnocellular-dorsal (where) and parvocellular-ventral (what) visual pathways but at distinct levels. The maintenance of space representation indicates the possible role of beta oscillations in the dorsal stream, while its role in the ventral pathway is proposed to be associated with the combination of object features. An interesting proposition made by this latest study envisions the role of beta oscillations as complementary to the function of alpha but with a distinctive feature of fast-updating spatial information through burst-like properties. Our results indicated SOA-dependent modulations in different frequency bands, further suggesting a complicated relationship between oscillatory phase and temporal organization of vision even when the SOA values are smaller than 100 ms (see also Purushothaman et al. 2000).

Delta-theta rhythms in visual processing

It has been proposed that oscillations in the delta band play a pivotal role as a general oscillatory framework for sensory processing and essential instruments for active input selection at the primary sensory cortex level (Schroeder and Lakatos 2009). With a focus on visual processing, Lakatos et al. (2008) revealed response gain modulations related to the phase of pre-stimulus delta oscillations. Previous studies documented performance improvements in contrast gain (Cravo et al. 2013) and faster reaction times (Stefanics et al. 2010) with increased delta (~1–4 Hz) phase synchronization. Similarly, previous psychophysical studies showed that visual detection performance fluctuates rhythmically around 4 Hz (Landau and Fries 2012; Landau et al. 2015). There are also studies pointing to a relationship between low-frequency oscillations (< 8 Hz) and the detection and discrimination of a visual feature (Bertrand et al. 2018, 2019). Interestingly, theta band has been associated with temporal segregation/integration performance (Ronconi et al. 2017; Ronconi and Melcher 2017) and spatial binding, including contour integration and detection (Stonkus et al. 2016). In particular, Stonkus et al. (2016) indicated a causal role of theta oscillations in contour processing. Our study provides further evidence by revealing an important link between contour discrimination performance and the phase of delta-theta oscillations over occipital and parieto-occipital areas. The long-lasting synchronized parietal-occipital theta oscillations, observed under conditions of maintained visibility, may reflect an uninterrupted feedforward flow of information and representation of target, and may preserve contour discrimination ability. According to Schroeder and Lakatos (2009), the sustained mode of delta-theta phase synchronization in our results may represent the preferred state of the system due to its processing efficiency and flexibility. Such rhythmic operation may emerge during task engagement and focus of attention, making brain rhythms ideal instruments for sensory selection. When the high-excitability phase of oscillations aligns with task-relevant sensory input, optimal processing of that input occurs. Thus, the phase of ongoing oscillations may serve as an indicator of perceptual cycles, with stimuli appearing at the optimal phase being optimally registered and perceived (Busch et al. 2009).

Conclusion

To conclude, our findings revealed important insights into the relationship between the post-stimulus phase of cortical oscillations and the dynamics of perceptual processing. We identified two clusters in (i) the alpha-beta band (10–25 Hz range) associated with the inhibitory mechanisms of masking, and (ii) the delta-theta band (2–5 Hz) reflecting preserved target visibility and contour discrimination. These findings are in agreement with the idea that the phase of neural oscillations reflects and predicts modulations of perceived visibility under different masking conditions, providing evidence for the critical role of ongoing oscillations in the temporal organization of perception.

Acknowledgments

We are grateful to the reviewers whose constructive and helpful comments significantly improved our manuscript.

Author contributions

Irem Akdogan, (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Visualization, Writing—original draft), Haluk Ogmen (Conceptualization, Supervision, Validation, Visualization, Writing—review & editing), Hulusi Kafaligonul (Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Validation, Visualization, Writing—review & editing).

Funding

This work was supported by the Scientific and Technological Research Council of Türkiye (119K368, BIDEB 2211-A).

Conflict of interest statement: The authors declare no competing financial interests.

Data availability

The dataset and analysis codes of the current study will be available upon request.