JMIR Publications

ABSTRACT

Background:

Developmental delay is a disorder in which a child does not grasp developmental milestones in areas like: motor skills, cognition, Speech, and societal interaction at an anticipated age. These delays include cognitive, speech and language, motor, and societal and emotional delays commonly caused by aspects like genetic circumstances, premature birth or low birth weight, neurological conditions, and environmental reasons.

Objective:

In this study, we used the Meta Quest Pro headset to generate an immersive virtual reality environment through task-based interactions. Including microexpressions, muscle activities, and eye-tracking data, were collected as time series data, apprehending dynamic facial gesture fluctuations.

Methods:

Subsequent data preprocessing, the primary analysis path leveraged statistical methods, which include: Kruskal–Wallis test, ReliefF, ANOVA, and minimum redundancy maximum relevance (MRMR), to extract and select the furthermost relevant features. The secondary path employed a transformer for automated feature extraction, then both feature sets were afterward classified using machine learning classifiers to evaluate their effectiveness.

Results:

Using the MRMR method, we achieved 86.4% accuracy with the Coarse Tree classifier with a ratio of 80:20, and 95.6% accuracy with quadratic support vector machine (SVM) with a ratio of 60:40. Whereas the transformer-based approach achieved an accuracy of 86.4% with Quadratic SVM and 95.5% with the Fine KNN classifier at the 60:40 split.

Conclusions:

This study contributes to the advancement of affordable and accessible screening tools for children worldwide.