Multi-Class Deep Learning Model for Detecting Pediatric Distal Forearm Fractures Based on the AO/OTA Classification.

Area of Science:

• Radiology
• Artificial Intelligence in Medicine
• Pediatric Orthopedics

Background:

• Accurate detection of pediatric distal forearm fractures is crucial for timely clinical treatment.
• The AO Foundation/Orthopaedic Trauma Association (AO/ATO) classification system provides a standardized framework for pediatric fractures.

Purpose of the Study:

• To develop a multi-class convolutional neural network (CNN) model for identifying pediatric distal forearm fractures.
• To classify fractures based on the AO/ATO pediatric fracture classification system.

Main Methods:

• Utilized the GRAZPEDWRI-DX dataset of wrist X-ray images (2008-2018).
• Trained a YOLOv4-based CNN object detection model on 7006 images from 1809 patients.
• Classified fractures into four types: FRM, FUM, FRE, and FUE, according to the AO/ATO system.

Main Results:

• The CNN model achieved high mean average precision on the validation set (0.92-0.97).
• On the test set, the model demonstrated strong performance with sensitivities (0.71-0.89), specificities (0.88-0.98), and AUCs (0.83-0.94).
• The model's performance (mean AUC 0.892) was comparable to a radiologist (0.922) and superior to an orthopedist (0.830).

Conclusions:

• The developed multi-class CNN model effectively identifies pediatric distal forearm fractures based on the AO/ATO classification.
• This AI-driven approach shows promise in supporting clinical decision-making for pediatric fracture management.