Fracture detection in pediatric wrist trauma X-ray images using YOLOv8 algorithm

Area of Science:

• Medical Imaging
• Computer Vision
• Orthopedics

Background:

• Hospital emergency departments manage numerous bone fracture cases, with pediatric wrist trauma being prevalent.
• Accurate interpretation of X-ray images by surgeons and radiologists is crucial for diagnosis and treatment planning.
• Current interpretation methods require extensive specialized training and can be time-consuming.

Purpose of the Study:

• To improve the performance of the YOLOv8 deep learning algorithm for detecting pediatric wrist trauma fractures.
• To leverage data augmentation techniques to enhance fracture detection accuracy.
• To develop a practical application for assisting surgeons in diagnosing wrist fractures from X-ray images.

Main Methods:

• Utilized the GRAZPEDWRI-DX public dataset of pediatric wrist trauma X-ray images.
• Applied data augmentation techniques to the dataset.
• Implemented and evaluated the YOLOv8 algorithm, comparing it with improved YOLOv7 and the original YOLOv8.

Main Results:

• The enhanced YOLOv8 model achieved a state-of-the-art (SOTA) mean average precision (mAP 50) of 0.638.
• This performance surpassed the improved YOLOv7 (0.634) and the original YOLOv8 (0.636).
• The developed application, "Fracture Detection Using YOLOv8 App," is designed for clinical use.

Conclusions:

• The data-augmented YOLOv8 model demonstrates superior performance in detecting pediatric wrist fractures.
• The developed application can assist surgeons in diagnosis, reduce errors, and provide valuable surgical information.
• This approach offers a promising tool for improving the efficiency and accuracy of pediatric fracture diagnosis.