Fracture Detection in Wrist X-ray Images Using Deep Learning-Based Object Detection Models.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Hospitals, particularly emergency departments, manage numerous wrist fracture cases.
• Accurate diagnosis relies on medical imaging, patient records, and physical examination.
• Physician support is crucial for timely and correct fracture diagnosis.

Purpose of the Study:

• To implement deep learning for automated wrist fracture detection on X-ray images.
• To assist physicians in diagnosing wrist fractures, especially in emergency settings.
• To develop a novel, high-performance detection model.

Main Methods:

• Evaluated 20 deep learning object detection procedures using models like Faster R-CNN and RetinaNet on a wrist X-ray dataset.
• Developed five ensemble models to enhance detection accuracy.
• Created a unique ensemble model, 'wrist fracture detection-combo (WFD-C)'.

Main Results:

• The WFD-C model achieved the highest detection performance.
• Achieved an average precision (AP50) of 0.8639 with the WFD-C model.
• Compared 26 different fracture detection models.

Conclusions:

• Deep learning models, particularly ensemble approaches, show significant promise for wrist fracture detection.
• The developed WFD-C model offers a robust tool to support clinical diagnosis.
• This AI-driven approach can enhance efficiency in emergency medical services.