Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Global Context Modeling in YOLOv8 for Pediatric Wrist Fracture Detection

Rui-Yang Ju, Chun-Tse Chien, Chia-Min Lin, Jen-Shiun Chiang

TL;DR

This work targets pediatric wrist fracture detection from X-ray images and introduces YOLOv8+GC, which embeds Global Context (GC) blocks into the YOLOv8 neck to capture long-range information with minimal overhead. On the GRAZPEDWRI-DX dataset, the proposed method achieves a $mAP_{50}$ of $66.32\%$, surpassing the previous SOTA while maintaining a lightweight footprint of $43.85\text{M}$ parameters and fast inference at $7.9\text{ ms}$. The study includes an ablation showing consistent gains across model sizes, yet notes persistent challenges with underrepresented classes (e.g., bone anomaly and soft tissue) due to dataset imbalance. Overall, YOLOv8+GC provides an effective, efficient CAD aid for clinicians in interpreting pediatric wrist X-rays, with performance that approaches SOTA levels while keeping resource usage low; future gains depend on more diverse data for underrepresented categories.

Abstract

Children often suffer wrist injuries in daily life, while fracture injuring radiologists usually need to analyze and interpret X-ray images before surgical treatment by surgeons. The development of deep learning has enabled neural network models to work as computer-assisted diagnosis (CAD) tools to help doctors and experts in diagnosis. Since the YOLOv8 models have obtained the satisfactory success in object detection tasks, it has been applied to fracture detection. The Global Context (GC) block effectively models the global context in a lightweight way, and incorporating it into YOLOv8 can greatly improve the model performance. This paper proposes the YOLOv8+GC model for fracture detection, which is an improved version of the YOLOv8 model with the GC block. Experimental results demonstrate that compared to the original YOLOv8 model, the proposed YOLOv8-GC model increases the mean average precision calculated at intersection over union threshold of 0.5 (mAP 50) from 63.58% to 66.32% on the GRAZPEDWRI-DX dataset, achieving the state-of-the-art (SOTA) level. The implementation code for this work is available on GitHub at https://github.com/RuiyangJu/YOLOv8_Global_Context_Fracture_Detection.

Global Context Modeling in YOLOv8 for Pediatric Wrist Fracture Detection

TL;DR

This work targets pediatric wrist fracture detection from X-ray images and introduces YOLOv8+GC, which embeds Global Context (GC) blocks into the YOLOv8 neck to capture long-range information with minimal overhead. On the GRAZPEDWRI-DX dataset, the proposed method achieves a of , surpassing the previous SOTA while maintaining a lightweight footprint of parameters and fast inference at . The study includes an ablation showing consistent gains across model sizes, yet notes persistent challenges with underrepresented classes (e.g., bone anomaly and soft tissue) due to dataset imbalance. Overall, YOLOv8+GC provides an effective, efficient CAD aid for clinicians in interpreting pediatric wrist X-rays, with performance that approaches SOTA levels while keeping resource usage low; future gains depend on more diverse data for underrepresented categories.

Abstract

Children often suffer wrist injuries in daily life, while fracture injuring radiologists usually need to analyze and interpret X-ray images before surgical treatment by surgeons. The development of deep learning has enabled neural network models to work as computer-assisted diagnosis (CAD) tools to help doctors and experts in diagnosis. Since the YOLOv8 models have obtained the satisfactory success in object detection tasks, it has been applied to fracture detection. The Global Context (GC) block effectively models the global context in a lightweight way, and incorporating it into YOLOv8 can greatly improve the model performance. This paper proposes the YOLOv8+GC model for fracture detection, which is an improved version of the YOLOv8 model with the GC block. Experimental results demonstrate that compared to the original YOLOv8 model, the proposed YOLOv8-GC model increases the mean average precision calculated at intersection over union threshold of 0.5 (mAP 50) from 63.58% to 66.32% on the GRAZPEDWRI-DX dataset, achieving the state-of-the-art (SOTA) level. The implementation code for this work is available on GitHub at https://github.com/RuiyangJu/YOLOv8_Global_Context_Fracture_Detection.
Paper Structure (13 sections, 4 figures, 2 tables)

This paper contains 13 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Network Architecture
  5. Global Context Block
  6. Experiments
  7. Dataset and Evaluation Metrics
  8. Data Preparation
  9. Training
  10. Ablation Study
  11. Experimental Results
  12. Discussion
  13. Conclusion

Figures (4)

  • Figure 1: Comparison of different models on the GRAZPEDWRI-DX dataset for fracture detection with the input image size of 1024. In terms of mAP 50, the proposed model YOLOv8+GC outperforms the YOLOv8+ResCBAM model, which is the SOTA model.
  • Figure 2: The detailed illustration of the proposed YOLOv8+GC model architecture, including backbone, neck, and head. The GC block is designed to be added to the neck of the model architecture.
  • Figure 3: Visualization of the accuracy of predicting each class using YOLOv8+ResCBAM-L model, YOLOv9-E model, and the proposed model YOLOv8+GC-L on the GRAZPEDWRI-DX dataset with the input image size of 1024.
  • Figure 4: Examples of prediction results of the YOLOv8+ResCBAM-L model and the proposed model YOLOv8+GC-L model for fracture detection with the input image size of 1024.