[Paper Review] Fracture Detection in Pediatric Wrist Trauma X-ray Images Using YOLOv8 Algorithm

Rui‐Yang Ju, Weiming Cai|arXiv (Cornell University)|Apr 11, 2023

Artificial Intelligence in Healthcare and Education13 citations

TL;DR

The paper trains and augments YOLOv8 models on the GRAZPEDWRI-DX pediatric wrist X-ray dataset to detect fractures, achieving state-of-the-art mAP@50 and releasing an accompanying Fracture Detection YOLOv8 App for surgeons.

ABSTRACT

Hospital emergency departments frequently receive lots of bone fracture cases, with pediatric wrist trauma fracture accounting for the majority of them. Before pediatric surgeons perform surgery, they need to ask patients how the fracture occurred and analyze the fracture situation by interpreting X-ray images. The interpretation of X-ray images often requires a combination of techniques from radiologists and surgeons, which requires time-consuming specialized training. With the rise of deep learning in the field of computer vision, network models applying for fracture detection has become an important research topic. In this paper, we use data augmentation to improve the model performance of YOLOv8 algorithm (the latest version of You Only Look Once) on a pediatric wrist trauma X-ray dataset (GRAZPEDWRI-DX), which is a public dataset. The experimental results show that our model has reached the state-of-the-art (SOTA) mean average precision (mAP 50). Specifically, mAP 50 of our model is 0.638, which is significantly higher than the 0.634 and 0.636 of the improved YOLOv7 and original YOLOv8 models. To enable surgeons to use our model for fracture detection on pediatric wrist trauma X-ray images, we have designed the application "Fracture Detection Using YOLOv8 App" to assist surgeons in diagnosing fractures, reducing the probability of error analysis, and providing more useful information for surgery.

Motivation & Objective

• Motivate fracture detection in pediatric wrist X-ray imaging where radiologist access is limited.
• Apply YOLOv8 with data augmentation to improve detection performance on a public dataset.
• Compare YOLOv8 variants against YOLOv7 and improved YOLOv8 models to establish SOTA performance.
• Develop a user-friendly application to assist pediatric surgeons in real-time fracture diagnosis.

Proposed method

• Train YOLOv8 models of different sizes on GRAZPEDWRI-DX (20,327 images) with data augmentation that adjusts image contrast and brightness using OpenCV addWeighted.
• Adopt YOLOv8 architecture with CSP backbone, C2f module, SPPF, FP-PAN neck, decoupled head, and Anchor-Free detection.
• Use Task Aligned Assigner (TOOD) for positive sample selection and a multi-branch loss combining BCE for classification, Distribution Focal Loss, and CIoU for localization.
• Compare SGD vs Adam optimizers and evaluate different input sizes (640 and 1024) on RTX 3080Ti, with training epochs reduced via early stopping.
• Evaluate performance using IoU-based metrics and Precision-Recall curves, reporting mAP@50 and mAP@50-95 across classes.

Experimental results

Research questions

• RQ1 What is the performance of YOLOv8 on pediatric wrist fracture detection in the GRAZPEDWRI-DX dataset?
• RQ2 Does data augmentation improve YOLOv8 fracture detection performance on this dataset?
• RQ3 How does the proposed YOLOv8-based approach compare with YOLOv7 and its variants in terms of mAP@50 and other metrics?
• RQ4 Can the trained model be deployed in a practical application to assist pediatric surgeons?

Key findings

• The proposed YOLOv8-based method with data augmentation achieves mAP@50 of 0.638, outperforming improved YOLOv7 (0.634) and original YOLOv8 (0.636).
• Ablation shows data augmentation notably improves bone-anomaly detection from 0.11 to 0.169 mAP@50, boosting overall performance.
• Compared with several SOTA models, the proposed approach matches or exceeds performance on mAP@50 and shows competitive precision and recall (e.g., 0.734 precision, 0.592 recall for the best reported model in Table 7).
• An application named 'Fracture Detection Using YOLOv8 App' was developed for macOS to assist surgeons by predicting fractures on pediatric wrist X-rays.
• The study reports that YOLOv8 variants with their training method surpass YOLOv7 and CBAM/GAM-enhanced YOLOv7 in mAP@50 on the GRAZPEDWRI-DX dataset.