Behavioral observation and assessment protocol for language and social-emotional development study in children aged 0-6: the Chinese baby connectome project

Abstract

Background: The global rise in developmental delays underscores the critical need for a thorough understanding and timely interventions during early childhood. Addressing this issue, the Chinese Baby Connectome Project (CBCP)'s behavior branch is dedicated to examining language acquisition, social-emotional development, and environmental factors affecting Chinese children. The research framework is built around three primary objectives: developing a 0-6 Child Development Assessment Toolkit, implementing an Intelligent Coding System, and investigating environmental influence.

Methods: Utilizing an accelerated longitudinal design, the CBCP aims to enlist a minimum of 1000 typically developing Chinese children aged 0-6. The data collected in this branch constitutes parental questionnaires, behavioral assessments, and observational experiments to capture their developmental milestones and environmental influences holistically. The parental questionnaires will gauge children's developmental levels in language and social-emotional domains, alongside parental mental well-being, life events, parenting stress, parenting styles, and family relationships. Behavioral assessments will involve neurofunctional developmental evaluations using tools such as the Griffiths Development Scales and Wechsler Preschool and Primary Scale of Intelligence. Additionally, the assessments will encompass measuring children's executive functions (e.g., Head-Toe-Knee-Shoulder), social cognitive abilities (e.g., theory of mind), and language development (e.g., Early Chinese Vocabulary Test). A series of behavior observation. experiments will be conducted targeting children of different age groups, focusing primarily on aspects such as behavioral inhibition, compliance, self-control, and social-emotional regulation. To achieve the objectives, established international questionnaires will be adapted to suit local contexts and devise customized metrics for evaluating children's language and social-emotional development; deep learning algorithms will be developed in the observational experiments to enable automated behavioral analysis; and statistical models will be built to factor in various environmental variables to comprehensively outline developmental trajectories and relationships.

Discussion: This study's integration of diverse assessments and AI technology will offer a detailed analysis of early childhood development in China, particularly in the realms of language acquisition and social-emotional skills. The development of a comprehensive assessment toolkit and coding system will enhance our ability to understand and support the development of Chinese children, contributing significantly to the field of early childhood development research.

Trial registration: This study was registered with clinicaltrials.gov NCT05040542 on September 10, 2021.

Keywords: Artificial intelligence; Behavioral observation; Infant; Language development; Social-emotional development; Young children.

Fig. 1

Developing process of the measures

Fig. 2

Workflow of the intelligent coding system on joint attention Notes. The model segments the video into seconds, and then into frames (30 frames per second). It extracts facial features from each frame and generates an attention map by combining this information with the position of the head. Simultaneously, it processes the full frame to focus on head positions, head orientation, and attention features. Then, these features are further processed through a Long Short-Term Memory network to capture dynamic changes in continuous frames. Finally, the model modulates the generated feature maps to produce a final heatmap, which reflects the collective joint attention of people in the image. The intensity of the heatmap is adjusted based on whether the participant’s attention is directed inside or outside the frame

Fig. 3

Multi-view 3D reconstruction pipeline Notes. This figure illustrates a moment in the Strange Situation, showing the participant placing toys into a box that the administrator is carrying while the mother observes from her seat. Our state-of-the-art 3D reconstruction algorithm effectively analyzes video streams to produce precise representations of the 3D human body by employing a multi-view camera configuration. This reconstruction simplifies the creation of intricate motion descriptions, which accurately describe the observed behaviors depicted in the recordings. These descriptions, along with a predetermined behavioral coding scheme, are combined into a comprehensive dataset. This dataset is subsequently employed to train an extensive language model, empowering sophisticated analysis of behaviors and automated assessment of observed actions