Analyzing pediatric forearm X-rays for fracture analysis using machine learning.

Area of Science:

• Orthopedic Surgery
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Pediatric forearm fractures are common in emergency departments.
• Current treatment challenges include resource limitations for specialized pediatric care, leading to delays and complications.
• There's a need for tools to support non-specialized centers in managing these fractures.

Purpose of the Study:

• To develop a machine learning (ML) model for analyzing pediatric forearm fractures.
• To assist clinical centers lacking pediatric orthopedic surgical expertise.
• To improve the accuracy and efficiency of fracture diagnosis and treatment planning.

Main Methods:

• Curated and preprocessed X-ray scans from 1250 children.
• Fine-tuned ML models using self-supervised learning with a vision transformer backbone.
• Employed region of interest (ROI) detection to identify critical fracture areas for analysis.

Main Results:

• The ML model, fine-tuned with ROI identification, achieved high performance.
• Achieved a true-positive rate (TPR) of 0.79 and a true-negative rate (TNR) of 0.74.
• Demonstrated strong classification performance with AUROC of 0.81 and AUPR of 0.86 on test data.

Conclusions:

• Machine learning models are feasible for predicting appropriate treatment for pediatric forearm fractures.
• The developed algorithm shows potential as a decision-support tool for non-specialized orthopedic providers.
• Further improvements could enhance diagnostic and treatment assistance capabilities.