Pediatric Wrist Fracture Detection in X-rays via YOLOv10 Algorithm and Dual Label Assignment System

TL;DR

This work tackles pediatric wrist fracture detection in X-ray images, a clinically important task due to high fracture prevalence and potential growth plate complications. It evaluates YOLOv10 variants on the GRAZPEDWRI-DX dataset, leveraging a dual-label assignment system and architectural innovations (CIB, large kernels, PSA) to improve detection efficiency and accuracy. The results show that YOLOv10 variants outperform YOLOv9 on this dataset, with YOLOv10-M delivering the best balance (mAP@50-95 of 51.9% and fracture sensitivity of 92.5%), and YOLOv10-X achieving the top overall mAP@50. These findings establish a new benchmark for pediatric fracture detection on GRAZPEDWRI-DX and suggest practical potential for real-time clinical screening, supported by publicly available code for replication and extension.

Abstract

Wrist fractures are highly prevalent among children and can significantly impact their daily activities, such as attending school, participating in sports, and performing basic self-care tasks. If not treated properly, these fractures can result in chronic pain, reduced wrist functionality, and other long-term complications. Recently, advancements in object detection have shown promise in enhancing fracture detection, with systems achieving accuracy comparable to, or even surpassing, that of human radiologists. The YOLO series, in particular, has demonstrated notable success in this domain. This study is the first to provide a thorough evaluation of various YOLOv10 variants to assess their performance in detecting pediatric wrist fractures using the GRAZPEDWRI-DX dataset. It investigates how changes in model complexity, scaling the architecture, and implementing a dual-label assignment strategy can enhance detection performance. Experimental results indicate that our trained model achieved mean average precision (mAP@50-95) of 51.9\% surpassing the current YOLOv9 benchmark of 43.3\% on this dataset. This represents an improvement of 8.6\%. The implementation code is publicly available at https://github.com/ammarlodhi255/YOLOv10-Fracture-Detection

Figures (4)

• Figure 1: YOLOv10 Architecture depicting the input, backbone, neck, head, and output.
• Figure 2: Precision, recall, and PR curves of YOLOv10-M variant across increasing confidence scores.
• Figure 3: Loss curves for YOLOv10-M variant.
• Figure 4: Inferences made with YOLOv10-M variant.