Development of GCN-based deep learning model for early prediction of comprehensive gross motor performance assessment in toddler (Preprint)

Abstract

Accurate and timely assessment of children’s developmental status is crucial for early diagnosis and intervention. More accurate and automated developmental assessments are essential due to lack of trained healthcare providers and imprecise parental reporting.

We developed a two-stage model to assess gross motor behavior and integrated the results to determine the overall gross motor status of toddlers.

To assess gross motor development, we selected four behaviors from the K-DST(Korean Developmental Screening Test for Infants & Children). In the first stage, each behavior was evaluated separately using a GCN-based algorithm. The resulting probability values for each label were input into the second-stage model, the XGBoost algorithm, to predict the overall motor performance status. For interpretability, we used Grad-CAM to identify important moments and relevant body parts during the movement performance. The variable importance was assessed during the overall performance prediction to determine the movements that contributed the most to the overall developmental assessment.

Behavioral videos of four gross motor skills were collected from 147 children, totaling 2,395 videos. The area under the curve (AUC) score of the GCN model, the evaluation model for each behavior, was found to be 0·79–0·90. Keypoint-mapping Grad-CAM visualization identified important moments in each behavior and differences in important body parts. In the XGBoost model, the overall gross motor performance status prediction model for stage 2 had an AUC of 0·90. Among the four behaviors, “Go downstairs” significantly contributed to the overall developmental assessment.

Using movement videos of 18–35-month-olds, we developed objective and automated models to evaluate each behavior and assess each child’s overall gross motor performance. We identified the behaviors important for assessing gross motor performance and developed methods to recognize important moments and body parts while evaluating gross motor performance.