The Development of a Real-Time Artificial Intelligence System Using a Convolutional Neural Network for the Assessment...

Area of Science:

• Medical Technology
• Artificial Intelligence
• Pediatrics

Background:

• Infant head control is a crucial developmental milestone.
• Early identification of developmental delays is vital for timely intervention.
• The pull-to-sit test is a common clinical assessment for infant head control.

Purpose of the Study:

• To develop and evaluate a real-time artificial intelligence (AI) system for assessing infant head control.
• To utilize a convolutional neural network for analyzing infant movements during the traction response.
• To determine the accuracy of the AI system in classifying head control as good or poor.

Main Methods:

• Infants aged 3-6 months were enrolled, and videos of the pull-to-sit test were recorded.
• Videos were processed into image frames for supervised machine learning.
• An AI system was trained and tested on labeled video data to assess sensitivity, specificity, and detection rate.

Main Results:

• The study included 184 infant videos (110 training, 74 testing).
• The AI system achieved high specificity (98.4%) for identifying good head control.
• Sensitivity for detecting poor head control was limited at 50.0%, with a 100% detection rate.

Conclusions:

• The AI-based system demonstrates high specificity in identifying good infant head control during the pull-to-sit test.
• The system's limited sensitivity for poor head control suggests it is best used as a supportive tool.
• Further development may enhance the AI's capability to detect developmental variations more comprehensively.