Assessing the stability of evoked, induced, and passive MEG responses for repeat testing with optically pumped magnetometers

Abstract

Passive and task-based MEG responses have been extensively studied in clinical populations to identify signatures that could serve as markers for clinical diagnosis or to monitor progression of treatment. Establishing the reliability of these responses across repeat testing as a benchmark is therefore essential. Emerging MEG technology using optically pumped magnetometers (OPMs) promises a new era for MEG, enhancing both research capabilities and clinical applications. However, the test-retest reliability of various MEG responses measured by these new systems has not yet been characterised. In this study, we measured a range of neural responses to task and rest using a whole-head OPM-MEG system. We assessed the stability of these responses over time in five adult participants, each tested across five different days. Our findings indicated that the well-established face-sensitive M170 response shows reliable group amplitude and latency over time, with a standard deviation of only 1 ms in latency. We showed that induced responses were more variable than evoked. Passive MEG power (via movie-watching as a pseudo-resting state metric) particularly in the alpha band, demonstrated high consistency across sessions, aligning with conventional MEG literature. Our results demonstrate reliability of a range of MEG metrics and provide a benchmark for evaluating changes over successive recordings.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-12131-2.

Introduction

Magnetoencephalography (MEG) is a neuroimaging technique that measures the magnetic fields generated by neuronal currents in the brain, uniquely benefiting from excellent temporal resolution and high spatial resolution^1,2. Consequently, MEG has become an invaluable tool in research and continues to gain traction for clinical applications³. Recent advances in quantum sensing technologies have resulted in next-generation MEG systems based on optically pumped magnetometers (OPMs) that can be worn directly on the head in helmets^4,5. These OPM-MEG systems provide numerous advantages over conventional MEG instrumentation. They do not require cryogenic cooling, vastly reducing operational costs and reliance on non-renewable, environmentally damaging cryogens. As they operate close to body temperature, they can be placed closer to the scalp surface, maximizing sensitivity to neural signals⁶; further, their miniaturized, flexible design allows placement in varying sized helmets, easily adapting to smaller heads^5,7,8. Finally, these systems enable, for the first time, motion during MEG scanning sessions⁴. These distinct advantages suggest a future whereby OPM-MEG systems are widely distributed in both research and clinical settings, not only replacing conventional systems but also enabling wider uptake of MEG globally⁹.

As we take steps towards OPM-MEG as a research and clinical tool, an important consideration is the reliability of these measures over time. Reliability over repeat testing refers to the consistency of a measure when conducted over multiple occasions with identical conditions; reliability is a fundamental requirement for clinical measures to ensure observed effects are due to functional changes rather than measurement variability. Reliability of conventional MEG has been investigated previously for a range of responses ^[e.g.10–12, with studies finding high stability of measures over long periods of time. However, the unique ability to move during OPM acquisition leads to inconsistent artefacts that may reduce the test-retest reliability of OPM measures, and the magnetometer design (rather than the hardware and synthetic gradiometers typically constructed in conventional MEG systems) means the sensors are inherently more affected by environmental magnetic fields that vary over time. Thus, the reliability of an OPM-based system must first be established before the other benefits of OPM systems can be realized for clinical application.

One recent study showed test-retest reliability of functional connectivity measures using OPM-MEG during a movie-watching paradigm¹³. While this study holds promise for reliability over short periods of time, data were collected during the same session for each participant, and therefore did not incorporate the additional sources of error that accompany scanning across multiple sessions. One potential source of error is the helmet placement, which may differ between scans (as positioning in the cryogenic MEG systems can vary as well), resulting in slight differences in coverage and compounding of coregistration errors. This is particularly pertinent for new OPM-based systems due to the typically lower numbers of sensors incorporated into the current OPM-MEG systems and the higher spatial frequencies observed closer to the scalp surface¹⁴. A second potential source of error is the ambient environmental noise, which although largely controlled (see methods, below), can vary from day to day, especially in downtown city locations¹⁵.

Here we used two tasks to investigate the test-retest reliability of a range of neural responses measured by our OPM system in a central city hospital. The first task comprises a series of emotional faces and concentric moving circular gratings. From this, we examined the well-established M170 response from fusiform gyri to the presentation of faces, which is known to occur at a consistent latency and localization in adults^16,17 and induced gamma (30–80 Hz) Event Related Synchronization (ERS) and alpha (8–13 Hz) Event Related Desynchronization (ERD) to the circles stimulus^18,19. The second task was a movie-based “resting-state”, used to investigate passive MEG power across canonical frequency bands. Movies are used here to enable lifespan-compliance, as previous work has shown that children are able to better comply with scanning conditions with movies than traditional fixation cross resting-state^20,21. One common statistic used to assess reliability is the intraclass coefficient (ICC), which describes the degree to which multiple measurements yield consistent results. ICC values range from 0 to 1, and here we categorise poor reliability as ICC < 0.5, moderate as 0.5 < ICC < 0.75, good as 0.75 < ICC < 0.9 and excellent as ICC > 0.9. In addition, as our sample size here is small due to the novelty of OPM technology, we present the confidence intervals of the ICC to put context to our statistics throughout. Thus, this study investigates the reliability of a range of OPM-MEG signals across repeated testing in adult participants spanning multiple sessions.

Results

OPM data were obtained from five adult participants while they performed two passive visual tasks (designed for use with young children). The first task included the presentation of a series of images of emotional faces alternating with concentrically moving circles, known to elicit an M170 (170 ms) evoked response and induced visual gamma ERS/alpha ERD, respectively. Participants were also shown a movie-based “resting-state” video²⁰. The participants were scanned at approximately the same time of day for 5 days (across a 6-day window) to investigate the stability of the responses over multiple sessions. Head motion data were acquired for 23 of the 25 sessions (the loss of 2 due to equipment failure); all participants showed similar amounts of motion per second across the tasks (with the most head motion in the final task of the inscapes video, with an average of 1.2 mm movement per second). There were no significant effects of session on head motion (p = 0.70).

Evoked: M170 response to faces

OPM data were filtered 2–40 Hz, epoched around face trials and beamformed to relate sensor signals to neuronal sources. Timeseries were extracted by taking the first principal component from bilateral fusiform gyri to examine the M170 evoked responses to the emotional face stimuli, shown in Fig. 1. Participants exhibited highly consistent localisation and signal morphology across sessions. Note that one participant consistently showed a larger response in the left fusiform, while the other four participants had a greater amplitude response in the right fusiform. The average standard deviations in localisation (i.e. radii of ellipses in Fig. 1a) across the multiple sessions were 1.94 mm, 5.07 mm and 2.82 mm in sagittal, coronal, and axial planes, respectively. Figure 1b shows the average from the first principal component for all participants and sessions, with the session average overlaid in transparent colour (i.e., average over the five participants for each day). Figure 1c shows the session average M170 response. These show highly consistent responses, with M170 amplitude of 0.003 and latency of 0.1670.001 s. Figure 1d shows the average and standard deviation evoked timeseries for each participant, with violin plots depicting the M170 amplitude and latency for each run. The average standard deviation in amplitude and latency for individuals was and 0.013 and 0.005 s, respectively. ICC was calculated for the M170 response, with amplitude having an ICC of 0.80 [CI = 0.5, 0.97] and latency of 0.56 [CI = 0.18, 0.93]. However, the moderate ICC in latency is to be expected as ICC is a poor measure of test-retest reliability for responses that are expected to have low variability between participants (i.e. in healthy adult participants, we expect all participants to have very similar latency).

Fig. 1.

Evoked response to emotional faces. (a) localization of peak fusiform response for each participant, where ellipsoids represent standard deviation in localisation across 5 sessions (registered to MNI brain). Note that one participant had stronger activity in the left fusiform, while all others showed stronger right sided activation. (b) shows the group average response from each day (transparent), with the average over all sessions in darker red, showing a clear M170. The amplitude and latency of the M170 across days is shown in c. d) shows the average peak fusiform response and each day’s response, with transparent overlays indicating session-level variation for each of the five participants, with violin plots showing the individual session M170 amplitudes and latencies.

Induced: visual gamma and alpha ERD to moving circles

Broadband (1–100 Hz) timeseries were extracted from the first principal component within a visual cortex mask to assess the induced gamma band (30–80 Hz) ERS and from the voxel showing greatest reduction in power relative to baseline across the alpha band (8–13 Hz), to investigate the time-frequency response (TFR) across the circles trials, as shown in Fig. 2. Although induced responses were more heterogeneous between participants, the relative spectra show that all participants exhibited consistent response morphologies across the five sessions. The average gamma peak frequency was 58 Hz, and amplitude was 21.3%, across all sessions and participants, which is comparable with results in cryogenic MEG studies¹⁹. The average standard deviation of the gamma peak amplitude across the five participants was 6.7%. ICC was calculated for the peak gamma frequency and amplitude (i.e. the maximum from the relative spectra), with poor frequency reliability of 0.07 [CI = −0.14, 0.69] and moderate amplitude reliability of 0.64 [CI = 0.27, 0.94]. However, the poor peak frequency reliability is driven by the heterogeneity in the gamma response, with several participants showing consistently low amplitude responses, hindering peak identification. Three of these participants were confirmed to demonstrate a very consistent response in a single session in a conventional MEG system following the analysis of these data, as presented in the supplemental material. These data again demonstrate the known heterogeneity in visual gamma response^12,19.

Fig. 2.

Induced response to concentric moving circles. (a) Average time-frequency response (relative to a baseline of −0.5–0 s) and relative spectra during stimulation (0.1–0.9 s) from the first principal component in visual cortex. (b) Average time-frequency response (relative to the same baseline) and relative spectra during stimulation from the voxel with greatest reduction in power from baseline in the alpha band (8–13 Hz). The average is plotted in dark colour, with each session overlaid in transparent lines, to show the intra-individual reliability.

In Fig. 2b, the TFRs show the response from the voxel displaying the minimum relative alpha power during the stimulation. The relative spectra (the average over the stimulation period) are shown below, where the peak frequency and amplitude of the response were determined from the minimum of the spectra. The average ERD frequency was 13.4 Hz and amplitude was − 45.9%. The average standard deviation of the amplitude across participants was . Note that one participant showed an ERD that spanned the low beta band (15–25 Hz). Peak ERD was highly consistent across subjects, which inflated test re-test variance, therefore resulting in poor ICC of 0.42 [CI = 0.06, 0.89]. The peak amplitude exhibited moderate ICC (0.54) [CI = 0.16, 0.92], potentially due to the floor effects in the ERD response. However, the overall morphology of an individual’s response remained largely consistent across sessions.

Power in canonical frequency bands and aperiodic component during passive viewing task

Data from the passive viewing paradigm were broadband filtered (2–80 Hz), epoched into 5 s segments and beamformed to the centroids of the 90 AAL regions to investigate whole-brain power across canonical frequency bands. Relative power was calculated from spectral density (PSD) of the region time series in the frequency bands of interest (4–8 Hz, 8–13 Hz, 13–30 Hz, 30–50 Hz and 50–80 Hz for theta, alpha, beta, low gamma, and high gamma, respectively) divided by the integral of the PSD. Figure S4 shows the topography of the power results. As expected, theta () oscillations dominated frontal regions, alpha () oscillations dominated occipital and parietal (“mu”) regions, beta () oscillations showed strong power across central and parietal cortex and low and high gamma oscillations ( and ) were predominantly frontal and temporal, in accordance with conventional MEG and previous OPM-MEG passive viewing or resting-state literature^13,22. Power was consistent between sessions, as shown in Fig. 3. We see a strong relation between power in each region for each participant across multiple runs, with higher intra-participant correlation than between-participant correlation. This is shown, for example, between participants’ first and last run in Fig. 3a, where particularly strong correlation is observed for alpha oscillations ( for the group), which was expected as these are typically the highest relative power, and therefore highest signal-to-noise. ICC was used to investigate the reliability of the power measures across brain regions and frequency bands. Figure 3b shows the ICC across the cortex for each frequency band. We observe moderate to good ICC, with average across all regions of = 0.73 [SD = 0.11], = 0.88 [SD = 0.05], = 0.73 [SD = 0.18], = 0.80 [SD = 0.10], = 0.73 [SD = 0.09]. Finally, we assessed the stability of the aperiodic component of the power across the whole brain, shown in Fig. 3c. We find a highly reliable signal, with the offset having an ICC of 0.75 [CI = 0.42, 0.97] and slope (or exponent) having an ICC of 0.86 [CI = 0.62, 0.98].

Fig. 3.

Passive viewing task shows high reliability. In the top panel, columns show the results for canonical frequency bands. (a) shows correlation of relative power for each region between the first and last runs for each participant, with colours representing the different participants, and (b) shows brain plots of intraclass correlation across brain regions. (c) shows the results from the analysis of the aperiodic component, with the whole-brain average absolute power spectra, the aperiodic component and periodic component for each participant. Transparent traces show the reliability within each participant across sessions, and the solid line shows the average across sessions.

Discussion

This study demonstrates the reliability of three independent OPM-MEG measures across multiple separate sessions, showcasing consistent responses despite incorporating errors associated with sensor positioning, coregistration and multi-day variability. Our findings extend previous work on the test-retest reliability of OPM-MEG-derived functional connectomes by interrogating the robustness of several neuronal responses using this new technology over multiple separate sessions spanning multiple days. Further, we demonstrate that OPM-MEG can measure repeatable metrics in a realistic clinical scanning scenario, with a city-centre hospital system and short, lifespan compliant paradigms.

The face-sensitive M170 response in bilateral fusiform gyri showed consistent latency across repeat testing, with a standard deviation of only 1 ms across the group. The amplitude of the response exhibited greater variability, with a standard deviation of approximately . Inter-and intra-subject amplitude variation with evoked responses is consistently higher than that of latency variation due to background activity and signal-to-noise variability²³. The variability in amplitude could be attributed to the task itself, with repeat testing evoking changes in amplitude as familiarity increases²⁴ or measuring differences in the participants’ attention levels. However, this may also be attributed to varying signal-to-noise, with different days having varying levels of environmental noise that may impact the amplitude (as it is relative to a baseline), or helmet placement meaning the peak response is not sampled uniformly across sessions. Despite these differences, we measured a consistent latency and have characterised the confidence within which we may measure amplitude differences. These results demonstrate that this measure with OPM-MEG can be a reliable marker for face processing in the brain, suitable for both research and clinical applications.

The induced responses to visual circularly oscillating gratings were more variable across participants, though this variability was replicated in an additional recording of three of the same participants in a cryogenic MEG system (see Figure S3). Peak induced gamma frequency was unreliable, though this may be due to a multi-peak effect, which is known in the induced visual gamma literature²⁵. Further, the between-participant heterogeneity in this response has consistently been observed in human visual gamma studies^12,19 including our own with OPM-MEG²⁶. The amplitude was more consistent across sessions, meaning those with a strong response consistently demonstrated a strong response. Therefore, while the heterogeneity of this response suggests that its power for distinguishing between groups, at least in young adults, as assessed in this study, is lower (i.e. a large sample size would be needed to overcome the larger variance), there is still value in the strong within-subject consistency. This response therefore could still have considerable value for tracking longitudinal changes in individuals. The alpha ERD peak frequency and amplitude also showed poor-to-moderate test-retest reliability, potentially due to the similarity across participants. Nevertheless, response morphology remained consistent within individuals and all individuals demonstrated a strong ERD in all sessions.

The movie-based passive viewing paradigm revealed high reliability in power measures across canonical frequency bands. It is worth noting that we did not test delta power, as the sensitivity of OPMs (as magnetometers) to the delta frequency band would likely be lower and has not yet been validated against conventional MEG systems. Across the frequency bands we tested, the ICC values across the cortex were moderate-to-good. These results indicate that power, particularly in the alpha band, is highly consistent across sessions. Further, we assessed the stability of the aperiodic component of the MEG signal and found that this had high test-retest reliability for both the offset and slope in the signal. These findings reinforce the reliability of our recent work on age-related changes in the aperiodic signal²⁷ and its importance in the study of disorders which implicate imbalance between excitatory and inhibitory signals, which have been shown to be related to the aperiodic component of the MEG/EEG signal²⁸. It is particularly important that we found stable task-free responses, as these measures are critical for understanding baseline brain function and passive viewing paradigms (as an alternative to resting-state with a fixation cross) are particularly useful in young and/or clinical populations²¹ where task-based measures may be challenging.

Neurophysiological recordings are used extensively with clinical populations to help understand a diversity of brain-behaviour relations or underlying neural mechanisms, and potentially aid in diagnosis, prognosis, or monitoring. Although many case-controlled studies over the years have found significant differences in a range of these metrics between clinical and control groups, increasingly evidence shows that heterogeneity and symptom overlap across groups, including control groups, suggests that transdiagnostic approaches are more appropriate; neuroimaging markers are better related to symptom or behavioural dimensions rather than diagnostic categories ^[e.g.29–31. Thus, future applications of neurophysiological signals will likely place greater emphasis on monitoring for clinical trials, where objective measures of brain processing can contribute to assessing efficacy of interventions. Foundational to such applications is the establishment of the reliability of these measures. As OPMs are becoming increasingly popular due to their easier use, lower operational costs, and greater comfort for participants than cryogenic MEG, we believe that this study, helping establish reliability metrics across three distinct, and well-used, neurophysiological markers will facilitate their future use in clinical research.

While our study provides promising evidence for the reliability of OPM-MEG measures, several limitations must be acknowledged. First, the sample size was small and included only five young adult participants due to the novelty of OPM technology and access to equipment and time for the multiple testing sessions. Future studies should include a larger, more diverse sample size to validate these results, perhaps by including repeat testing on clinical and control cohorts. Additionally, although head motion was monitored, we did not directly examine the relations between motion and reliability metrics, as participants were generally quite still (though natural motion throughout would have exceeded cryogenic MEG requirements). The participants in this study were also scanned within a six-day window, which may not capture the variability over longer periods; this should be explored in future work. Second, our paradigms were designed for lifespan compliance, involving far fewer trials than previous adult-focused studies (e.g., 60 possible circles trials versus over 200^12,32,33) to accommodate the attention span and tolerance of very young children and clinical populations. Although this may impact the signal-to-noise we are able to obtain and therefore the data reliability, we believe it is critical for generalisability to clinically relevant protocols. With these protocols we were still able to elicit responses clearly in our task-based paradigms and detect power across frequency bands, which provides validation that these short paradigms are sufficient despite low numbers of trials. Third, it is worth noting that our data were analysed using pseudo-MRIs due to lack of anatomical MRI scans for all participants. While our recent work has shown this to be a reliable method that produces very similar results to individual anatomy³⁴ this inevitably produces an additional source of potential error, particularly in the localisation of responses between sessions as coregistration errors appear to be inflated by pseudo-MRI usage. This may reduce ICC values across the responses. We have mitigated these errors by using a coarse parcellation and inflating masks to account for individual differences. Finally, the reliability of the gamma response was lower, with two participants showing minimal gamma ERS. While this is comparable to previous literature¹⁹ and in the cryogenic MEG (Figure S3), future work in optimising experimental design and array configuration³⁵ may enable better characterisation of this heterogeneous response.

In conclusion, our study demonstrates the reliability of OPM-MEG technology for measuring a range of neural responses across multiple sessions. These findings support the potential of OPM-MEG as a robust tool for both research and clinical applications and can guide the use of this technology in practice.

Methods

Participants and paradigms

Five participants, aged 22 to 32 years (3 male), were scanned for five sessions over six days. The study protocols were approved by the Research Ethics Boards of the Hospital for Sick Children, all research was performed in accordance with the Declaration of Helsinki and written informed consent was obtained from all participants. Each participant’s scanning session took place at approximately the same time each day, within a 2-hour window, as outlined in Fig. 4c. Each session consisted of two paradigms, which are shown in Fig. 4a and comprise of the following:

Fig. 4.

Experimental design. (a) an overview of the experimental paradigms, with facial expressions from the NimStim Set of Facial Expressions³⁶ (All faces from the Nim Stim data set are publicly available for scientific research at https://macbrain.org/resources/) (b) diagram of the OPM system (c) outline of OPM-MEG sessions.

• “Faces Circles”: Participants viewed a sequence of visual stimuli, comprising facial expressions from the NimStim Set of Facial Expressions³⁶ and inwardly drifting circular gratings, presented in a random order. A total of 80 facial images displaying either “happy” or “angry” emotions were shown for 500 ms each. In addition, 60 instances of maximum contrast circular gratings with a visual angle of ~ 7° moving at 1.2°/s were presented for 1 s. Interspersed between the faces and circles, cartoon character ‘catch’ trials were presented, which are used to maintain attention for younger participants for whom this paradigm was designed; these were not analysed. Each trial was followed by a fixation cross with a jittered duration of 1250 ± 200 ms. This task took approximately 5 minutes.

• Passive viewing – “Inscapes”: Participants watched the initial five minutes of the Inscapes movie²⁰ which presents non-social, non-verbal moving abstract shapes accompanied by a continuous, gentle piano score.

OPM-MEG data collection

The OPM-MEG system (Cerca Magnetics Limited, Nottingham, UK) consisted of a 3D-printed helmet fitted with an evenly distributed array of 40 dual-axis sensors (QuSpin Inc., Colorado, USA), shown in Figure S1. This setup provided whole-head coverage with 80 channels operating at a sampling rate of 1200 Hz. Participants were seated wearing the helmet in a magnetically shielded room (MSR; Vacuumschmelze, Hanau, Germany) positioned between a pair of bi-planar coils and a reference array of OPM sensors, which dynamically compensated for background magnetic fields and drift over time^15,37. A diagram of the OPM system is shown in Fig. 4b. An optical tracking system with four infrared cameras (OptiTrack Flex 13, NaturalPoint Inc., Oregon, USA) mounted on the bi-planar coils and infrared pearl markers placed on the helmet was used to continuously track head movements. A 3D model of the participant’s head, with the helmet on, was obtained using an Einscan H 3D-scanner (Shining 3D, Hangzhou, China) to determine the position of the OPM sensors relative to the head^5,38. A 3-minute empty-room noise recording was acquired prior to each participant’s session.

Data analyses

Preprocessing OPM data

All data were preprocessed using a custom in-house OPM preprocessing pipeline in Matlab (MathWorks Inc., version R2024a, Massachusetts, USA) using the Fieldtrip toolbox³⁹, which has been used in previous OPM analyses^27,40. Briefly, noisy channels were removed using an outlier detection algorithm, which iteratively identified channels with a median PSD across 10-s epochs that were more than three median absolute deviations from the median across all channels using Matlab’s isoutlier function⁴⁰ and homogeneous field correction was used to suppress interference from outside sources (e.g. environmental noise⁴¹). Data were filtered broadband from 1 to 100 Hz and additional environmental noise was removed via notch filtering peaks that were identified from the empty room noise recording (power spectrum pre- and post- notch filtering are shown in Figure S2).

Coregistration and forward modelling

During the first scanning session, participants underwent a second 3D digitisation using the Einscan 3D scanner without the helmet on and with a swimming cap to flatten their hair to their scalp. From this, a pseudo-MRI was generated using FSL flirt and an age-matched template MRI^8,34,42. OPM data were coregistered via feature matching between the pseudo-MRI and the 3D digitisation of the participant wearing the helmet separately for each session^5,38. The forward model for source analysis was derived from the pseudo-MRI using a single shell model⁴³ in Fieldtrip toolbox across a 5 mm grid spanning the extracted brain from the warped template.

Evoked analyses

Pre-processed data were epoched into trials spanning -0.5–1 s around emotional face stimulus onset. Trials with excessive artefacts were removed using a threshold approach⁴⁰resulting in an average of ±trials. Data were analysed in source space using an LCMV beamformer⁴⁴. Here, data were filtered 2–40 Hz using a Butterworth filter and beamformed to the bilateral fusiform regions from the AAL atlas⁴⁵ within a 5 mm grid. Principal component analysis was applied to the timeseries from each fusiform voxel, and the first component was extracted. Compared to approaches that rely on a timeseries from a single voxel, this method captures the most significant variance in the signal across the fusiform region, providing a more representative measure of its activity. From this timeseries, the M170 was identified and manually flipped to ensure the peak was the same (negative) polarity for all participants and sessions. The localisation was derived by taking the voxel timeseries with maximum (absolute) correlation with the first principal component. The peak amplitude and latency for each session were identified from the minimum between 140 and 190 ms following stimulus onset. Intraclass correlation coefficient was calculated for latency and amplitude using the Pingouin statistical package in Python.

Induced analyses

For induced analyses, pre-processed OPM data were epoched into trials spanning  -1–2 s around the onset of the concentric moving circles stimuli. Trials with excessive artefacts were again removed using the same thresholding approach as evoked analyses, resulting an average of ± trials. For induced gamma analyses, broadband data were beamformed (using an LCMV beamformer⁴⁴) to a visual mask, which was derived from inflating the 5 mm grid spanning the left and right cuneus from the AAL atlas (i.e., regions 45 and 46⁴⁵) using a 5 mm spherical structuring element²⁶. The first principal component was extracted from the mask time series for time-frequency analyses. The time-frequency representation (TFR) of oscillatory activity was computed using multitaper convolution implemented in MNE-Python⁴⁶. Power was estimated for frequencies from 4 to 80 Hz in 1 Hz steps to capture gamma band activity. The number of cycles per frequency was set to half the frequency to optimize spectral concentration while preserving temporal precision. Baseline correction was applied using a percent change approach relative to an average pre-stimulus baseline (-0.5 to -0.1 s; the baseline was reduced to avoid edge artefacts). Relative spectra were computed by taking the average of the time-frequency response in the gamma range (30–80 Hz) during the stimulation period (0.1 to 0.9 s) to assess the spectral features per session. For the alpha/low beta analysis, data were filtered to the alpha band (8–13 Hz) and beamformed to the same visual mask. The voxel with the greatest reduction in alpha power during this window was identified, from which a broadband timeseries was extracted. The TFR across the low frequencies was again computed using multitaper methods in MNE-Python, spanning 4 Hz to 80 Hz in 1 Hz steps, with the same number of cycles. TFRs were again baselined using a relative percent change approach from the average pre-stimulus (-0.5 to 0 s) and the relative spectra were calculated using the same methods as for gamma, but spanning lower frequencies (4–30 Hz). Peak frequency was assessed by taking the frequency at which the maximum occurred in the relative spectra for the gamma response, and minimum for the alpha response. The amplitude was taken as the height of the maximum/minimum from the relative spectra. Intraclass correlation coefficient was calculated for peak frequency and amplitude using the Pingouin statistical package in Python.

Passive viewing task analyses

Passive viewing “Inscapes” movie data were epoched into 5 s trials. Noisy trials were identified and removed using the same threshold, resulting in an average of 60 ± trials. Data filtered to 2–80 Hz were beamformed (using an LCMV beamformer⁴⁴) to 90 AAL regions, and timeseries for each region were extracted from the region centroid. From this, the PSD was computed using Welch’s method and normalised by the integral across the whole PSD. The relative power in each frequency band was obtained by taking the integral of the normalised PSD across the frequency window of interest. Correlation between power in each region for each frequency band between the first and last sessions was investigated using Pearson correlation. Intraclass correlation coefficient was calculated for each brain region for each frequency band using the Pingouin statistical package in Python. Finally, we assessed the aperiodic component of the signal across all 90 regions using the specparam package in Python⁴⁷ and a pipeline previously described and published by our lab²⁷. Briefly, the PSDs from 2 to 40 Hz were input into the model without a knee or bend with no restrictions on the number of peaks or peak widths. For each PSD, the slope (also known as exponent and describes the steepness of exponential decay) and the offset (the uniform shift of the spectrum across frequencies) were extracted and averaged across brain regions for each participant and session. The ICC was assessed on the whole-brain metric using Pignouin statistical package in Python.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

N.R. – Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Visualization, Writing – original draft. J.B. – Data curation, Formal analysis, Investigation, Project administration, Software, Writing – original draft. M.M.W – Conceptualization, Methodology, Software, Writing – reviewing & editing. S.C.C - Methodology, Software, Writing - reviewing & editing.M.J.T. – Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing – reviewing & editing.

Funding

Funding was provided by the Canadian Institutes of Health Research (PJT-178370) and the Simons Foundation Autism Research Initiative (SFARI; 2021 Human Cognitive and Behavioural Science award).

Data availability

The datasets generated and/or analysed during the current study are not publicly available due to ongoing analyses but are available from the corresponding author on reasonable request. The analysis code is available on GitHub: https://github.com/nsrhodes/opm_retest.

Declarations

Competing interests

The authors declare no competing interests.