Abstract
Diagnosing and treating abnormalities in the wrist, specifically distal radius and ulna fractures, is a crucial concern among children, adolescents, and young adults, with a higher incidence rate during puberty. However, the scarcity of radiologists and the lack of specialized training among medical professionals pose a significant risk to patient care.
This study employs state-of-the-art single-stage deep neural network-based detection models YOLOv5, YOLOv6, YOLOv7, and YOLOv8 to detect wrist abnormalities. Through extensive experimentation, we found that these YOLO models outperform the commonly used two-stage detection algorithm, Faster R-CNN, in fracture detection.
YOLOv8m demonstrated the highest fracture detection sensitivity of 0.92 and mean average precision (mAP) of 0.95. Meanwhile, YOLOv8x recorded the highest mAP of 0.77 for all classes on the GRAZPEDWRI-DX pediatric wrist dataset, highlighting the potential of single-stage models for enhancing pediatric wrist imaging.
System Demo
Interactive Web Interface: Our system provides a user-friendly interface for uploading pediatric wrist X-ray images and receiving automated fracture detection results
Clinical Impact:
- Addresses Radiologist Scarcity: Provides automated screening to support medical professionals in areas with limited specialist access
- Reduces Diagnosis Time: Instant analysis of pediatric wrist X-rays with high accuracy detection
- Mobile-Ready: Available as both web application and Android mobile app for offline usage
- High Precision: Achieves fracture detection mAP of 0.95 with YOLOv8x model
Key Research Contributions
Novel Finding: First comprehensive study demonstrating that single-stage YOLO models significantly outperform traditional two-stage detection methods (Faster R-CNN) for pediatric wrist fracture detection.
📊 Performance Benchmark
Comprehensive evaluation of 23 different detection procedures on 1016 test samples from GRAZPEDWRI-DX dataset
🎯 State-of-the-Art Results
YOLOv8x achieved 0.77 mAP for all classes and 0.95 mAP specifically for fracture detection
🏥 Clinical Application
End-to-end system deployment with web interface and Android mobile application for real-world usage
🔬 Model Comparison
First systematic comparison of YOLOv5, YOLOv6, YOLOv7, and YOLOv8 variants for medical imaging applications
Experimental Results
Dataset: GRAZPEDWRI-DX pediatric wrist dataset with 1016 test samples. All models evaluated using precision, recall, and mean average precision (mAP) metrics.
YOLOv8 Model Performance
| Model Variant | Image Size (pixels) | Precision | Recall | mAP@0.5 | mAP@0.5:0.95 | Model Weights |
|---|---|---|---|---|---|---|
| YOLOv8n | 640 | 0.73 | 0.58 | 0.59 | 0.36 | Download |
| YOLOv8s | 640 | 0.72 | 0.63 | 0.65 | 0.39 | Download |
| YOLOv8m | 640 | 0.60 | 0.92 | 0.56 | 0.36 | Download |
| YOLOv8l | 640 | 0.74 | 0.60 | 0.62 | 0.41 | Download |
| YOLOv8x | 640 | 0.79 | 0.64 | 0.77 | 0.53 | Download |
Model Performance Comparison: Mean Average Precision (mAP) scores for all YOLO variants and Faster R-CNN across fracture detection and all classes
System Architecture
End-to-End System Pipeline: From X-ray image input through YOLO-based detection to clinical diagnosis output
Implementation Features:
🌐 Web Application
Flask-based web interface for easy image upload and result visualization
📱 Mobile App
Android application for offline diagnosis in resource-limited settings
🔄 Real-time Processing
Instant fracture detection with bounding box visualization
💾 Model Management
Support for multiple YOLO model variants with easy switching
Technical Specifications:
- Input: Pediatric wrist X-ray images (DICOM, PNG, JPG formats)
- Processing: YOLO-based object detection with confidence scoring
- Output: Fracture localization with bounding boxes and confidence levels
- Deployment: Containerized deployment with Docker support
Clinical Impact & Applications
Addressing Healthcare Challenges:
Our system directly addresses the critical shortage of radiologists and specialized training in pediatric fracture diagnosis, particularly important given the higher incidence of distal radius and ulna fractures during puberty.
⚡ Rapid Diagnosis
Reduces diagnosis time from hours to seconds, enabling immediate clinical decision-making
🎯 High Accuracy
95% mAP for fracture detection ensures reliable clinical support and reduced false negatives
🌍 Global Accessibility
Mobile app enables deployment in resource-limited settings without internet connectivity
📊 Clinical Support
Assists medical professionals with consistent, objective analysis of pediatric wrist X-rays
Future Clinical Integration: The system is designed for integration into existing radiology workflows, providing decision support rather than replacement of clinical expertise.
Installation & Usage
Quick Start: Get the system running locally in minutes with our comprehensive setup guide.
Installation Steps:
- Clone the repository:
git clone https://github.com/ammarlodhi255/pediatric_wrist_abnormality_detection-end-to-end-implementation.git - Navigate to project directory:
cd pediatric_wrist_abnormality_detection-end-to-end-implementation - Create virtual environment:
python3 -m venv env - Activate environment:
source env/bin/activate - Install dependencies:
pip install -r requirements.txt - Run the application:
flask run
Contact & Acknowledgments
Research Team
Ammar Ahmed -
ammarlodhi68@gmail.com
LinkedIn
Abdul Manaf -
abdulmanafsahito@gmail.com
LinkedIn
Ali Shariq Imran - NTNU Gjøvik, Norway
Institutions
Sukkur IBA University - Pakistan
NTNU Gjøvik - Norway
Linnaeus University - Sweden
Published in: Biomedical Signal Processing and Control
Acknowledgments
We express our sincere gratitude to Dr. Sher Muhammad Daudpota, Zenun Kastrati, and all supervisors who provided guidance throughout this research. Special thanks to the committee members and the institutions that supported this work.
This research contributes to advancing automated medical imaging analysis for improved pediatric healthcare outcomes.