This protocol describes how synchronized electroencephalography, electrocardiography, and behavioral recordings were captured from infant-caregiver dyads in a home setting.
Prior hyperscanning studies that record the brain activities of caregivers and children concurrently have primarily been conducted within the confines of the laboratory, thus limiting the generalizability of results to real-life settings. Here, a comprehensive protocol for capturing synchronized electroencephalography (EEG), electrocardiography (ECG), and behavioral recordings from infant-caregiver dyads during various interactive tasks at home is proposed. This protocol demonstrates how to synchronize the different data streams and report EEG data retention rates and quality checks. Additionally, critical issues and possible solutions with respect to the experimental setup, tasks, and data collection in home settings are discussed. The protocol is not limited to infant-caregiver dyads but can be applied to various dyadic constellations. Overall, we demonstrate the flexibility of EEG hyperscanning setups, which allow experiments to be conducted outside of the laboratory to capture participants’ brain activities in more ecologically valid environmental settings. Yet, movement and other types of artifacts still constrain the experimental tasks that can be performed in the home setting.
With the simultaneous recording of brain activities from two or more interacting subjects, also known as hyperscanning, it has become possible to elucidate the neural basis of social interactions in their complex, bidirectional, and fast-paced dynamics1. This technique has shifted the focus from studying individuals in isolated, tightly controlled settings to examining more naturalistic interactions, such as parent-child interactions during free play2,3, puzzle-solving4, and cooperative computer games5,6. These studies demonstrate that brain activities synchronize during social interactions, i.e., show temporal similarities, a phenomenon termed interpersonal neural synchrony (INS). However, the great majority of hyperscanning studies have been confined to laboratory settings. While this allows for better experimental control, it may come at the expense of losing some ecological validity. Behaviors observed in the laboratory may not be representative of the participants' typical everyday interactive behaviors due to the unfamiliar and artificial setting and the nature of the tasks imposed7.
Recent advances in mobile neuroimaging devices, such as electroencephalography (EEG) or functional near-infrared spectroscopy (fNIRS), alleviate these issues by removing the requirement for participants to remain physically connected to the recording computer. Thus, they allow us to measure participants' brain activities while they interact freely in the classroom or in their homes8,9. The advantage of EEG compared to other neuroimaging techniques, such as fNIRS, is that it has an excellent temporal resolution, which makes it particularly suitable for investigating fast-paced social dynamics10. Yet, it comes with the caveat that the EEG signal is highly vulnerable to motion and other physiological and non-physiological artifacts11.
Despite this, the first studies have successfully implemented EEG hyperscanning set-ups in realistic environments and conditions. For instance, Dikker et al.12 measured the EEG signal of a group of students while they engaged in various classroom activities, including attending lectures, watching videos, and participating in group discussions. This study, along with other studies8,9, has predominantly utilized dry EEG electrodes to ease the process of conducting measurements in non-laboratory settings. Compared to wet electrodes, which require the application of conductive gel or paste, dry electrodes offer notable advantages in terms of usability. They have been shown to exhibit comparable performance to wet electrodes in adult populations and stationary conditions; however, their performance may decrease in motion-related scenarios due to increased impedance levels13.
Here, we present a working protocol to capture synchronized recordings from a low-density seven-channel liquid gel EEG system with a single lead electrocardiography (ECG) connected to the same wireless amplifier (sampling rate: 500 Hz) of infant-caregiver dyads in a home setting. While active electrodes were used for adults, passive electrodes were used instead for infants since the latter typically comes in the form of ring electrodes, thereby easing the process of gel application. Additionally, EEG-ECG recordings were synchronized to three cameras and microphones to capture the participants' behaviors from different angles. In the study, 8-12-month-old infants and their caregivers engaged in a reading and play task while their EEG, ECG, and behaviors were recorded. To minimize the impact of excessive movement on EEG signal quality, the tasks were conducted in a table-top setting (e.g., utilizing the kitchen table and an infant highchair), requiring participants to remain seated throughout the interaction task. Caregivers were provided with three age-appropriate books and table-top toys (equipped with suction cups to prevent them from falling). They were instructed to read to their child for approximately 5 min, followed by a 10 min play session with the toys.
This protocol details the methods for collecting synchronized EEG-ECG, video, and audio data during the reading and play tasks. The overall procedure, however, is not specific to this research design but is appropriate for different populations (e.g., parent-child dyads, friend dyads) and experimental tasks. The method of synchronization of different data streams will be presented. Further, a basic EEG preprocessing pipeline based on Dikker et al.12 will be outlined, and EEG data retention rates and quality control metrics will be reported. Since the specific analytical choices depend on a variety of factors (such as task design, research questions, EEG montage), hyperscanning-EEG analysis will not be detailed further, but instead, the reader will be referred to existing guidelines and toolboxes (e.g.,14 for guidelines;15,16 for hyperscanning analysis toolboxes). Finally, the protocol discusses challenges and potential solutions for EEG-ECG hyperscanning in the home and other real-world settings.
The protocol described has been approved by the Institutional Review Board (IRB) of the Nanyang Technological University, Singapore. Informed consent was obtained from all adult participants and from parents on behalf of their infants.
1. Considerations of equipment and space at home sessions
2. Preparations before the session
3. Experiment preparation at the participant's home
Figure 1: Top-down view of set up. (1) Infant-facing camcorder. (2) Combined view camcorder. (3) Caregiver-facing camcorder. Please click here to view a larger version of this figure.
4. EEG and ECG sensor application for the caregiver
5. EEG and ECG sensor application for the infant
6. Creating a trigger box for multi-modal data synchronization
NOTE: Since different sensor data streams (i.e., EEG, ECG, video, and audio) will start recording at different time points during the session, they need to be manually synchronized to create a single timeline of events. Thus, a common event is needed that can be captured by both the camcorder (i.e., LED light) and the amplifier (i.e., digital or analog signal). To achieve this, an in-house synchronization trigger box is used, which can be built using a simple microcontroller unit program, as detailed below.
Figure 2: Building of Trigger Box. (A) Microcontroller circuit diagram for trigger box; (B) Interiors of the built trigger box; (C) Trigger box connected to the adult and infant EEG-ECG amplifiers, the trigger push button, and the power bank. Please click here to view a larger version of this figure.
Figure 3: High active and low active trigger port settings. Depending on the initial state of the trigger pin (0 or 1), the trigger port setting (High Active, HA or Low Active, LA) is chosen so that the marker is produced at the end of the pulse (when the trigger push button is released). Please click here to view a larger version of this figure.
7. Sensor streams synchronization
8. Parent-infant interaction experiment
9. Clearing up at the end of the experiment
10. Data quality assurance
11. Data processing
Participants included in this study were 8- to 12-month-old, typically developing infants and their mother and/or grandmother who spoke English or English and a second language at home. The 7-electrode EEGs and a single-lead ECG of adults and infants, as well as video and audio recordings from three cameras and microphones, were acquired simultaneously during the tasks. Neural activities were measured over F3, F4, C3, Cz, C4, P3, and P4 according to the international 10-20 system. The different data streams were temporally aligned and cut at the beginning and end of the experiment so that all recordings started at timepoint t = 0 (Figure 4).
Figure 4: Synchronization of data streams. Three cameras (infant-view, combined-view, and caregiver-view), caregiver and infant raw ECG, as well as caregiver and infant raw EEG, are synchronized to the same timeline. Please click here to view a larger version of this figure.
EEG data retention rates and quality metrics for the first 5 dyads of the data set with a total of 10 participants are presented in Table 1. After bad channel rejection (Figure 5), data segments containing artifacts were rejected using an automated voltage threshold followed by visual inspection of the remaining segments (Figure 6). Results showed that EEG recordings had an average length of M = 562.96 s (SD ± 148.94 s). From these, M = 34.30% (SD ± 13.00%) of the adult data and M = 46.32% (SD ± 16.63%) of the infant data were accepted following automatic and manual rejection. If considering only matched data between adult and infant, the retention rate dropped to M = 20.58% (SD ± 9.51%), leaving M = 215.00 s (SD ± 117.54 s) of matched task data. Further, a total of 0 to 2 channels per dyad were excluded due to consistently poor data quality.
Figure 5: Identifying bad channels. An infant EEG data scroll and Power Spectral Density (PSD) plot for 7 EEG channels in which either channel Cz is observed to be a flatline (A: data scroll, B: PSD plot), or channel F3 is excessively noisy (C: data scroll, D: PSD plot). Bad channel detection was performed in EEGLAB17. Please click here to view a larger version of this figure.
Figure 6: Artifact rejection. Epochs with artifacts were rejected automatically according to a (A) rejection threshold, (B) followed by manual rejection through visual inspection. The highlighted portion of the plot shows rejected segments according to the rejection threshold (A) or manual rejection (B), respectively. Data were visualized in EEGLAB17. Please click here to view a larger version of this figure.
Adult | Child | Matched | |||||
ID | Recording length (s) | Bad channels | % accepted epochs | Bad channels | % accepted epochs | Bad channels | % accepted epochs (matched) |
1 | 898 | NA | 35.7 | Cz | 25.2 | Cz | 15.3 |
2 | 1234 | NA | 38.2 | Cz, F3 | 61.8 | Cz, F3 | 21.2 |
3 | 1088 | F3, F4 | 52.4 | F3, F4 | 63.1 | F3, F4 | 36.7 |
4 | 873 | NA | 27.9 | P3 | 34.6 | P3 | 12.8 |
5 | 975 | NA | 17.2 | NA | 47.0 | NA | 16.9 |
Table 1. EEG data quality report for 5 dyads during the experimental tasks.
In this protocol, we conduct measurements in the participants' homes where infants and caregivers may feel more comfortable and their behaviors may be more representative of their real-life interactions as opposed to a laboratory setting, thus, increasing ecological validity7. Further, recordings in the home environment may ease the burden on the participants, e.g., with respect to travel times, and may thus make certain participant groups more accessible. However, along with these advantages, naturalistic EEG hyperscanning recordings in real-world contexts pose their own set of challenges and limitations with regard to experimental design and protocol as well as data artifacts. In the following, challenges and possible solutions for home recordings are discussed.
The naturalistic environment may introduce a set of confounding variables such as space, temperature, and interruptions, which may differ between participant groups at home but stay constant in a controlled laboratory setting. The EEG hyperscanning protocol requires a lot of technical equipment, e.g., several cameras, microphones, and recording laptops, and therefore, lack of sufficient space in participant homes may sometimes be an issue. Researchers must be aware not to set up equipment haphazardly or somewhere surrounded by clutter. For example, it is important to be mindful not to set up devices on tables with food or drink items and to make sure camera tripods are not blocking the way in narrow spaces. One way to prevent issues with space would be to visit the participant's home beforehand to appropriately plan ahead of time for any space constraints. It is also helpful to send reminders to the participants to have the required space cleared of items. Cameras and tripods should be placed out of the way as much as possible, especially when out of reach from where the infant is sitting during the session. Most of all, all parties' safety must be considered at all stages of the set-up. Another factor that researchers may encounter in naturalistic settings is varying temperatures. In Singapore, where temperatures are high throughout the day and year, sweat artifacts may occur in the EEG data, which can be better controlled in the laboratory environment with appropriate air conditioning. Using fans to keep participants cool also introduces other artifacts due to having electrical appliances in proximity, and the blowing air may move participants' hair, as well as the EEG wires, resulting in poor data quality. Ideally, air conditioning should be used during the session as it will keep participants cool. Still, if this is not possible, an overhead fan or standing fan may be used instead while making sure it is not placed too close to the participants to avoid creating noise in the EEG data. Other alternatives would be to schedule the session during a cooler time of the day if possible so that sweat artifacts can be avoided. Finally, researchers also need to be wary that interruptions may occur in a naturalistic setting, especially if conducting the session at participant's homes. Family members may be in the vicinity, which can cause a violation of privacy when filming the session in a common room where they may be walking by. It can also be a distraction for the infant to see other caregivers or family members during the task, which may bias the EEG measurements. It would be best to remind participants that for the session to run smoothly, it would be ideal to have other family members in a different room. Researchers can also try to conduct the session as efficiently as possible so as not to inconvenience the other members of the household too much. Lastly, researchers must ensure that all data is collected and that the necessary items are completed before leaving the participant's home. Having a clear and organized checklist of documents and items to be completed can help avoid missing any important steps and also help complete them efficiently and in a timely manner.
Apart from the confounding variables found in a naturalistic environment, there are also some aspects of the protocol that will need to be adjusted for each session in a natural setting that are otherwise controlled in a laboratory environment. Standardization will not be possible for certain aspects, such as camera angles and lighting. Flexibility in set-up while also ensuring high-quality and comparable data is crucial. Camera angles may change with each participant's home as a result of differences in the layout and space, which may make it more difficult for later annotations of videos for specific events and behavioral metrics. Similarly, the lighting will also differ in each home, which can affect the quality of the video. Researchers can be adequately prepared by creating a general set of standards that can be adapted, such as making sure the participants are not seated against a main source of light and knowing what camera angles to prioritize. Another varying factor would be the furniture available to use in each session. Since researchers most likely cannot bring furniture to the participants' homes, they will have to rely on furniture that the participants already have. Due to this, the different furniture used can change the physical dynamic between the caregiver and the infant. For example, various kinds of baby chairs will change the height and position at which the infant is seated during the task. This may affect the way the caregiver interacts with the child and also affect the EEG data due to potential muscle movement artifacts or other factors. During the preprocessing stage of data analysis, researchers may be able to identify the EEG artifacts caused by specific movements by seeking guidance from the synchronized videos. Furthermore, having a general idea of what kinds of behaviors are going to be observed or analyzed can help to ensure that the necessary data is captured despite the varying physical dynamics.
A further implication of the home-environment naturalistic setup of EEG experiments concerns the quality and usability of physiological sensor data. EEG recordings are prone to artifact interference from environmental (non-physiological, such as line noise18) and physiological sources (ocular, sweat, myogenic)19. Although wireless EEG is generally less vulnerable to line noise, electrical devices in the home, e.g., fans, TV screens, and aircon, will introduce noise artifacts. Movement artifacts, on the other hand, are even more prominent in a naturalistic setting and contribute to lower data retention11,20, reduction in signal-to-noise ratio21, and vulnerability in data analysis in interpretation11. Dyadic EEG and infant EEG present an additional challenge in data retention due to lower recording durations, less stereotypical artifact presentations, and, in the case of hyperscanning, the necessity of clean analyzable segments to be matched in time14,22,23. Mitigation of these factors relies on thoughtful experimental design and well-calibrated experimental setup22. Although high-density EEG compositions allow for some artifact correction and data augmentation techniques, such as independent component analysis (ICA) removal of canonical noise components, this is not recommended with low-density setups. In contrast, relying on hand annotation of artifacts and removal of affected EEG channels and segments leads to greater data loss. The proposed protocol can also be performed with more EEG channels but at the cost of a longer preparation time. These advantages of shorter acquisition time versus richer EEG data must be carefully weighed against each other. Here, a realistic estimate of data retention rates from the naturalistic home recordings is reported, adhering to strict quality standards using a combination of automated voltage spikes labeling and manual artifact rejection. Although the retention rates were low (M = 34% for adults and M = 46% for infants), they are within the excepted range for naturalistic infant-adult EEG recordings, e.g., as a comparison, Dikker et al.12 reported a retention rate of 38% during the discussion task in adult EEG using dry electrodes. The amount of clean data recovered from the paradigm can be fed into further analyses, such as time-frequency-based connectivity analyses. Alternative semi-automated pipelines for artifact correction of low-density EEG recordings (e.g., HAPPILEE24), albeit out of the scope of the current paper, may help remove artifacts without the use of ICA and thus significantly reduce data loss.
To ensure high-quality EEG but feasible data collection, researchers will need to consider how the naturalistic setting affects the tasks that are chosen for the experimental session. For example, the choice of tasks can be based on what would be commonly found in participant homes, such as a dining table, chairs, baby chairs, playmat, etc. This would allow for less bulky equipment or furniture that needs to be transported back and forth and would also reduce the set-up and clean-up time. In this experiment, books and toys that are suitable for tabletop play were used, allowing caregiver and child to maintain a naturalistic play dynamic while also limiting free movement so that muscle movement EEG artifacts can be reduced. As a result, in the current protocol, the toys were chosen based on what would reflect natural interactions. For example, toys with suction that can be placed in a stationary position for the caregiver and child to engage with on the table have the advantage that they cannot fall off the table, which may cause motion artifacts when the caregiver tries to pick them up. Researchers also need to be wary of preparation and clean-up time to reduce participant burden.
Although choosing to conduct EEG hyperscanning measurements in a naturalistic environment has many benefits for more ecologically valid data, researchers should be aware of the limitations and challenges that may arise from the experimental design and implement sufficient steps to mitigate the effects as much as possible. Researchers must strive to strike a balance between an ecological design and experimental control when optimizing their paradigm and planning their visits. As described above, some flexibility with respect to the experimental set-up is needed, which, however, introduces more variability between participants. While this is undesirable from an experimental perspective, it may be more reflective of the participants' real-world environments. Additionally, the naturalistic setup may introduce more and other types of artifacts to the EEG data, as discussed above. These can, to some extent, be mitigated by appropriate EEG preprocessing and analysis techniques but generally can lead to a higher loss and lower quality of data. Further, the equipment used, in particular the cameras and tripods, comes with the disadvantages of being relatively bulky and heavy, thus making it difficult to transport and less suitable for confined spaces. Finally, the wet electrode system needs additional experimental materials (e.g., gel, syringes, gloves, wipes) and longer preparation times. Experimenters must be very careful not to leave a mess in the participants' homes, e.g., getting gel on parts of the furniture, and explain in advance that there is a risk that the infant may do so. Dry electrodes can be a good alternative to circumvent these issues and save set-up time. Thus, for hyperscanning recordings in larger groups (e.g., classrooms), these may be the method of choice (e.g., see 12). Therefore, by refining and adapting this protocol to the circumstances at hand, it has the potential to be applied in many different types of naturalistic settings, such as schools and workplaces, to capture a larger variety of hyperscanning and behavioral data.
The authors have nothing to disclose.
The work was funded by a Presidential Postdoctoral Fellowship Grant from Nanyang Technological University that was awarded to VR.
10 cc Luer Lock Tip syringe without Needle | Terumo Corporation | ||
actiCAP slim 8-channel electrode set (LiveAMP8) | Brain Products GmbH | ||
Arduino Software (IDE) | Arduino | Arduino IDE 1.8.19 | The software used to write the code for the Arduino microcontroller. Alternate programming software may be used to accompany the chosen microcontroller unit. |
Arduino Uno board | Arduino | Used for building the circuit of the trigger box. Alternate microcontroller boards may be used. | |
BNC connectors | BNC connectors to connect the various parts of the trigger box setup. | ||
BNC Push button | Brain Products GmbH | BP-345-9000 | BNC trigger push button to send triggers. |
BNC to 2.5 mm jack trigger cable (80 cm) | Brain Products GmbH | BP-245-1200 | BNC cables connecting the 2 LiveAmps to the trigger box. |
BrainVision Analyzer Version 2.2.0.7383 | Brain Products GmbH | EEG analysis software. | |
BrainVision Recorder License with dongle | Brain Products GmbH | S-BP-170-3000 | |
BrainVision Recorder Version 1.23.0003 | Brain Products GmbH | EEG recording software. | |
Custom 8Ch LiveAmp Cap passive (infant EEG caps) | Brain Products GmbH | LC-X6-SAHS-44, LC-X6-SAHS-46, LC-X6-SAHS-48 | For infant head sizes 44, 46, 48 . Alternate EEG caps may be used. |
Dell Latitude 3520 Laptops | Dell | Two laptops, one for adult EEG recording and one for infant EEG recording. Alternate computers may be used. | |
Dental Irrigation Syringes | |||
LiveAmp 8-CH wireless amplifier | BrainProducts GmbH | BP-200-3020 | Two LiveAmps, one for adult EEG and one for infant EEG. Alternate amplifier may be used. |
Manfrotto MT190X3 Tripod with 128RC Micro Fluid Video Head | Manfrotto | MT190X3 | Alternate tripods may be used. |
Matlab Software | The MathWorks, Inc. | R2023a | Alternate analysis and presentation software may be used. |
Power bank (10000 mAh) | Philips | DLP6715NB/69 | Alternate power banks may be used. |
Raw EEG caps | EASYCAP GmbH | For Adult head sizes 52, 54, 56, 58. Alternate EEG caps may be used. | |
Rode Wireless Go II Single Set | Røde Microphones | Alternate microphones may be used. | |
Sony FDR-AX700 Camcorder | Sony | FDR-AX700 | Alternate camcorders or webcams may be used. |
SuperVisc High-Viscosity Gel | EASYCAP GmbH | NS-7907 |