Related Experiment Video
Updated: Jul 9, 2026
06:59
Fracture Apparatus Design and Protocol Optimization for Closed-stabilized Fractures in Rodents
Published on: August 14, 2018
Pediatric fracture classification in plain radiographs using EfficientNetV2 with proximal policy optimization
Malek Barhoush1, Ruba Khasawneh2, Salem Alhatamleh3
1Information Technology Department, Cybersecurity Program, Faculty of Information Technology & Computer Sciences, Yarmouk University, Irbid, Jordan.
Scientific Reports
|July 7, 2026
Summary
This study introduces an AI framework for detecting pediatric fractures in X-rays, improving accuracy by using EfficientNetV2-Small and Proximal Policy Optimization (PPO) to reduce missed fractures in children.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Pediatric Radiology
Background:
- Pediatric fracture detection in radiographs is challenging due to developmental bone variations.
- High inter-reader variability and missed fracture rates necessitate automated decision support.
Purpose of the Study:
- To develop and evaluate a binary classification framework for automated pediatric fracture detection.
- To improve diagnostic accuracy and reduce the risk of overconfident missed fractures.
Main Methods:
- A two-stage framework combining EfficientNetV2-Small for feature extraction and Proximal Policy Optimization (PPO) for policy refinement.
- Supervised warm-up with cross-entropy loss followed by PPO fine-tuning using a confidence-aware asymmetric reward function.
- Contrast Limited Adaptive Histogram Equalization preprocessing on the KAUH Pediatric Fracture X-ray Dataset (1221 radiographs).
Main Results:
- The proposed model achieved 91.86% test accuracy, 95.90% AUC, and 98.38% sensitivity.
- Outperformed baseline Convolutional Neural Network (CNN) models including MobileNetV2, ResNet50, Xception, DenseNet121, and base EfficientNetV2-Small.
- The PPO refinement specifically targeted high-confidence misclassifications to mitigate the risk of missed fractures.
Conclusions:
- The developed AI framework shows significant promise for enhancing pediatric fracture detection accuracy.
- The confidence-aware reward function in PPO effectively addresses the clinical need to reduce overconfident missed fractures.
- Further multi-center validation is required before clinical deployment of this research prototype.
Related Concept Videos
01:27Fractures: Bone Repair
Treatment for a fracture is based on the type of break, the bone affected, and the patient's age.
Minor fractures with no bone displacement are treated by immobilizing the fractured bone using a cast or splint. However, in the case of fractures with displaced bones, the broken bones are repositioned before immobilization to ensure successful healing without deformation and loss of function. The realignment of fractured bone ends is performed through a process called reduction. If the procedure...
Minor fractures with no bone displacement are treated by immobilizing the fractured bone using a cast or splint. However, in the case of fractures with displaced bones, the broken bones are repositioned before immobilization to ensure successful healing without deformation and loss of function. The realignment of fractured bone ends is performed through a process called reduction. If the procedure...
01:18Classification of Bones
The bones of the human skeletal system are of varied shapes, sizes, and functions. They can be classified based on their shape and function into four major classes: long bones, short bones, flat bones, and irregular bones. Some classifications include a fifth type, the sesamoid bones, as a separate class, whereas others categorize them under short bones.
Long and Short Bones
The appendicular skeleton, particularly the upper and lower limbs, is primarily made of long and short bones. The long...
Long and Short Bones
The appendicular skeleton, particularly the upper and lower limbs, is primarily made of long and short bones. The long...