Neural Dynamics of Parent-Infant Interaction: How 2 Different Processing Approaches Can Shape Our Understanding

Parent-infant interaction is a cornerstone of early social development. So it’s important to understand the neuro-physiological processes that occur during these interactions. In recent years, this became more accessible through advancements in hyperscanning techniques.

Hyperscanning is a simultaneous recording of brain activity in multiple participants. It’s emerged as a powerful tool to explore Interpersonal Neural Synchronization (INS). INS is the temporal alignment of neural activity between interactive partners.

In recent years, EEG has been widely adopted to study dyadic INS – parent-infant interaction. But, the field faces a significant challenge: the lack of standardization in EEG data pre-processing.

While there are guidelines, discrepancies in pre-processing routines can greatly impact the results. So how do methodological choices in pre-processing affect the interpretation of neural synchrony?

Authors of this study address the gap by comparing automated EEG pre-processing pipeline with a manual approach. By focusing on both pre-processing outcomes and INS estimates, they aim to shed light on this important topic.

Collecting data and defining the pipelines: automatic vs. manual

31 pairs of mother-infant dyads participated in the study. They were enrolled in a Face-to-Face Still-Face (FFSF) procedure. These are the sessions in the procedure:
1. 2 minutes of face-to-face play interaction (Play)

2. 1 minute of maternal still face (Still-Face)
3. 2-minute face-to-face playing (Reunion)

FFSF procedure is a standard which allows researchers to observe infants’ socio-emotional competence and ability to cope with stress.

During these sessions, EEG is recorded on both parent and infant using the Smarting PRO system.

Let’s take a look at the defined pipeline flow in the study:

Authors used the Neonatal EEG Artifact Removal (NEAR) algorithm to detect bad channels. That includes flat and outlier channels. The difference is, in the manual pipeline the coder had to visually check the results. In the end, they can either accept or deny the automatic detection.

Another step that differed between pipelines is the detection of artifactual components. To decompose the data into independent components (ICs), authors used the Infomax ICA algorithm. The resulting components can have different sources – brain but also heart, eye-movements, muscles etc.

ICs are rejected completely automatically in the first pipeline. But, in the manual pipeline the coder removed them by assessing the following metrics:

• • Check the ICs flagged by ADJUST algorithm – see more details on this link.
  • Manually check the remaining ICs and reject those displaying a probability higher than 50% of containing ocular artifacts.

Lastly, removing bad epochs with high-voltage values is automatic or manual with visual inspection.

Data quality assessment and computing INS indices

Authors wanted to ensure good data quality before going into any analysis. They computed Signal-to-Noise ratio (SNR) for the dyadic pairs. Any components in the data over 100Hz are considered white noise:

Positive SNR values show a higher signal power compared to noise. All the infants’ and the majority of the parents’ signals showcased a SNR that exceeded 10 dB, meaning that the dataset was adequate for a further computation.

Phase-Locking Value (PLV) as a measure of parent-infant interaction

PLV is a measure of frequency-specific transients of phase locking independent of amplitude. Explaining why it’s been widely used in Hyperscanning studies.

For PLV calculation, there’s two frequency bands that are of interest:

1. Theta band which is associated with changes in direct gaze during joint/individual play.
2. Alpha band related to changes in emotional states during interaction.

The channels used in PLV calculation are below:

The final metrics consists of averaged frontal and central PLV values per session (Play, Reunion), and frequency band.

[Results] Differences in Pre-Processing Outputs

The automated pipeline preserved more of the original signal, rejecting less ICs and epochs. But let’s take a look at the difference in component labelling:

It seems that the automated pipeline contained more noise-related ICs – non-brain components. In contrast, the manual pipeline identified more ocular artifacts and rejected more ICs and epochs. These differences may stem from variations in channel interpolation between pipelines. Also, the sensitivity of artifact detection tools. Especially in infant data, where validation of tools is less robust.

These findings highlight the impact of pre-processing choices on signal quality and artifact removal.

Significant changes in PLV levels between processing pipelines

Theta-band PLVs are significantly higher after automated pre-processing compared to manual, in frontal and central regions. On the other hand, alpha-band showed no significant differences between the pipelines.

Theta PLVs might be influenced by residual ocular artifacts in the automated pipeline. Another reason is stricter rejection of theta-related brain activity in the manual pipeline. These results suggest that phase-based INS measures are sensitive to the degree of automation.

Authors do note that further research could determine which approach results in more accurate estimates.

Towards automated pre-processing for parent-infant interaction EEG data

Automated pre-processing offers advantages in speed and reproducibility. This makes it a promising option for large-scale dyadic EEG studies.

Yet, its impact on signal quality and INS estimates requires further investigation. Future research should explore the effects of different pre-processing parameters and thresholds. As well as their influence on amplitude-based INS measures.

Standardizing pre-processing methods in developmental EEG hyperscanning could enhance comparability across studies. It would help in advancing our understanding of parent-infant interaction.

Read the entire study here.