Fracture Detection in Pediatric Wrist Trauma X-ray Images Using YOLOv8 Algorithm

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 train YOLOv8 (the latest version of You Only Look Once) model on the GRAZPEDWRI-DX dataset, and use data augmentation to improve the model performance. The experimental results show that our model have reached the state-of-the-art (SOTA) real-time model performance. Specifically, compared to YOLOv8s models, the mean average precision (mAP 50) of our models improve from 0.604 and 0.625 to 0.612 and 0.631 at the input image size of 640 and 1024, respectively. 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.

If you find our paper useful in your research, please consider citing:

@article{ju2023fracture,
  title={Fracture detection in pediatric wrist trauma X-ray images using YOLOv8 algorithm},
  author={Ju, Rui-Yang and Cai, Weiming},
  journal={Scientific Reports},
  volume={13},
  number={1},
  pages={20077},
  year={2023},
  publisher={Nature Publishing Group UK London}
}

• Linux (Ubuntu)
• Python = 3.9
• Pytorch = 1.13.1
• NVIDIA GPU + CUDA CuDNN

  pip install -r requirements.txt

• GRAZPEDWRI-DX Dataset (Download Link)

• Download dataset and put images and annotatation into ./GRAZPEDWRI-DX_dataset/data/images, ./GRAZPEDWRI-DX_dataset/data/labels.

    python split.py
  

• The dataset is divided into training, validation, and testing set (70-20-10 %) according to the key patient_id stored in dataset.csv.

• The script then will move the files into the relative folder as it is represented here below.

   GRAZPEDWRI-DX_dataset
      └── data   
           ├── meta.yaml
           ├── images
           │    ├── train
           │    │    ├── train_img1.png
           │    │    └── ...
           │    ├── valid
           │    │    ├── valid_img1.png
           │    │    └── ...
           │    └── test
           │         ├── test_img1.png
           │         └── ...
           └── labels
                ├── train
                │    ├── train_annotation1.txt
                │    └── ...
                ├── valid
                │    ├── valid_annotation1.txt
                │    └── ...
                └── test
                     ├── test_annotation1.txt
                     └── ...
  

The script will create 3 files: train_data.csv, valid_data.csv, and test_data.csv with the same structure of dataset.csv.

• Data augmentation of the training set using the addWeighted function doubles the size of the training set.

  python imgaug.py --input_img /path/to/input/train/ --output_img /path/to/output/train/ --input_label /path/to/input/labels/ --output_label /path/to/output/labels/

For example:

  python imgaug.py --input_img ./GRAZPEDWRI-DX/data/images/train/ --output_img ./GRAZPEDWRI-DX/data/images/train_aug/ --input_label ./GRAZPEDWRI-DX/data/labels/train/ --output_label ./GRAZPEDWRI-DX/data/labels/train_aug/

• The path of the processed file is shown below:

   GRAZPEDWRI-DX_dataset
      └── data   
           ├── meta.yaml
           ├── images
           │    ├── train
           │    │    ├── train_img1.png
           │    │    └── ...
           │    ├── train_aug
           │    │    ├── train_aug_img1.png
           │    │    └── ...
           │    ├── valid
           │    │    ├── valid_img1.png
           │    │    └── ...
           │    └── test
           │         ├── test_img1.png
           │         └── ...
           └── labels
                ├── train
                │    ├── train_annotation1.txt
                │    └── ...
                ├── train_aug
                │    ├── train_aug_annotation1.txt
                │    └── ...
                ├── valid
                │    ├── valid_annotation1.txt
                │    └── ...
                └── test
                     ├── test_annotation1.txt
                     └── ...
  

You can get the open source code of YOLOv8 through YOLOv8 official GitHub.

Before training the model, make sure the path to the data in the meta.yaml file is correct.

• meta.yaml

  # patch: /path/to/GRAZPEDWRI-DX/data
  path: 'E:/GRAZPEDWRI-DX/data'
  train: 'images/train_aug'
  val: 'images/valid'
  test: 'images/test'

• Arguments

• Example

  cd Bone_Fracture_Detection_YOLOv8
  yolo train model=yolov8n.pt data=./GRAZPEDWRI-DX/data/meta.yaml epochs=100 batch=16 imgsz=640 save=True device=0 workers=4 pretrained=yolov8n.pt optimizer=SGD lr0=0.01

Use gdown to download the trained model from our GitHub:

  gdown https://github.com/RuiyangJu/Bone_Fracture_Detection_YOLOv8/releases/download/Trained_model/best.pt

• Arguments

• CLI

  yolo val model=/path/to/best.pt data=/path/to/meta.yaml

• CLI

  yolo predict model=/path/to/best.pt source=/path/to/image

• Example

  yolo predict model=best.pt source='./test.png'

For research project agreement, we don't release APP code, please refer to YOLOv7 Bone Fracture Detection and our paper for details.