Automated Wrist Ultrasound Image Bone Enhancement and Segmentation Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Orthopedics

Background:

• Wrist fractures are common pediatric injuries, typically diagnosed with X-rays.
• Ultrasound offers a radiation-free, rapid alternative for evaluating suspected fractures.
• Accurate segmentation of bony structures in ultrasound is crucial for severity assessment.

Purpose of the Study:

• To develop and evaluate an AI-based segmentation framework for pediatric wrist ultrasound images.
• To assess the impact of Contrast-Limited Adaptive Histogram Equalization (CLAHE) on segmentation accuracy.
• To explore the clinical feasibility of automated segmentation in pediatric emergency settings.

Main Methods:

• Utilized the nnU-Net model for semantic segmentation of epiphysis, metaphysis, and carpal bones.
• Applied Contrast-Limited Adaptive Histogram Equalization (CLAHE) as a preprocessing step.
• Conducted experiments on a dataset of 16,865 training and 3,822 testing ultrasound images.

Main Results:

• The nnU-Net framework achieved a DICE score of 0.874 with CLAHE preprocessing.
• Segmentation performance was slightly improved with CLAHE (0.874) compared to without (0.872).
• Demonstrated the feasibility of automated segmentation by lightly trained users.

Conclusions:

• AI-powered segmentation of wrist ultrasound images is feasible for pediatric fracture assessment.
• CLAHE image enhancement improves the accuracy of AI-based segmentation.
• This approach can serve as a triage tool, potentially reducing the need for X-rays in pediatric emergency care.