Automated Neonatal Hip Ultrasound System for Diagnosing Developmental Dysplasia of Hips Using Assistive AI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Orthopedics

Background:

• Developmental dysplasia of the hip (DDH) requires accurate early diagnosis.
• Infant hip ultrasonography is a key diagnostic tool.
• AI integration can enhance diagnostic capabilities.

Purpose of the Study:

• To develop and evaluate an AI-based system for diagnosing DDH using infant hip ultrasonography.
• To automate the Graf algorithm for DDH diagnosis.
• To assess the performance of various AI models in DDH detection.

Main Methods:

• Utilized the Graf algorithm to create an automated DDH diagnostic model.
• Evaluated multiple AI architectures including NASNetMobile, MobileNetV1, DenseNet121, EfficientNetV2B0, ResNet50, UnestedUNet, and DeepLabV3Plus.
• Integrated AI with a handheld ultrasound device and smartphone for portable diagnostics.

Main Results:

• The AI model achieved an average Graf angle error rate of 0.21 compared to experts.
• NASNetMobile showed the highest Area Under the Curve (AUC) of 0.864.
• UnestedUNet demonstrated high segmentation accuracy (Dice coefficient 0.794) with efficient runtime.
• DeepLabV3Plus showed robust segmentation performance on a portable device.

Conclusions:

• AI integration significantly enhances the accuracy and efficiency of DDH diagnosis.
• Portable AI-powered ultrasound devices offer accessible diagnostic solutions for DDH.
• The developed AI system shows transformative potential in pediatric orthopedic diagnostics.