Tebyan: An AI-powered system for estimating developmental levels from children's human figure drawings

Area of Science:

• Artificial Intelligence
• Developmental Psychology
• Pediatric Health

Background:

• Early identification of developmental concerns is crucial for timely intervention.
• The Draw-A-Person (DAP) test is a common tool for assessing child development.
• Objective assessment of DAP test results can be challenging.

Purpose of the Study:

• To develop and evaluate Tebyan, an AI-powered mobile application for estimating children's developmental levels from DAP drawings.
• To support the early identification of developmental concerns through automated analysis.
• To compare AI-estimated developmental age with chronological age to flag potential issues.

Main Methods:

• Nine deep learning models (MobileNet, ResNet, EfficientNet) were trained using binary, four-class, and eight-class configurations.
• Models predicted developmental age from DAP drawings.
• Performance was evaluated using accuracy, sensitivity, specificity, precision, and F1-scores.

Main Results:

• Two-class models achieved approximately 80% accuracy, while four-class and eight-class models showed moderate (55-65%) and low (30-40%) performance, respectively.
• Model performance decreased with increased class granularity.
• The four-class MobileNet model was selected for Tebyan due to its balance and stability.

Conclusions:

• Tebyan offers an AI-driven method for estimating developmental levels from children's drawings.
• The application compares predicted developmental age with chronological age, highlighting potential areas for further evaluation.
• While not a diagnostic tool, Tebyan serves as a supportive resource for caregivers and educators.