Explainable AI for early developmental disability detection: a neuro-fuzzy approach

Area of Science:

• Pediatric Health
• Artificial Intelligence in Medicine
• Developmental Neuroscience

Background:

• Developmental disabilities impact 1 in 6 children, with diagnosis delays disproportionately affecting underserved populations.
• Traditional screening methods lack the transparency needed for clinical integration.
• Early identification and intervention are crucial for improving outcomes in children with developmental disabilities.

Purpose of the Study:

• To develop and validate an interpretable AI model for early screening of developmental disabilities.
• To assess the diagnostic accuracy and clinical utility of the Adaptive Neuro-Fuzzy Inference System (ANFIS).
• To address the limitations of traditional machine learning in clinical settings.

Main Methods:

• An Adaptive Neuro-Fuzzy Inference System (ANFIS) was developed using a longitudinal dataset of 5,000 children (ages 1-6).
• The model incorporated Gaussian membership functions, Sugeno fuzzy inference, and age-adjusted ratios.
• Performance was evaluated using standard classification metrics and five-fold cross-validation.

Main Results:

• Significant differences in cognitive, behavioral, motor, and social interaction scores were observed between diagnostic groups (p < 0.001).
• The ANFIS model achieved 96.0% accuracy, 87.5% sensitivity, and 97.6% specificity on a test set.
• Cross-validation yielded a mean accuracy of 89.2% ± 3.4%, with Cognitive_Social_Ratio being the most influential indicator.

Conclusions:

• The ANFIS approach offers a clinically relevant balance of accuracy and interpretability for decision support.
• This AI tool shows promise as a viable clinical decision-support system for developmental disabilities.
• Further validation in diverse populations is recommended before widespread clinical implementation.