The GRAZPEDWRI-DX is a open dataset containing 20327 annotated pediatric trauma wrist radiograph images of 6091 patients, treated at the Department for Pediatric Surgery of the University Hospital Graz between 2008 and 2018. Several pediatric radiologists annotated the images by placing bounding boxes to mark 9 different classes:
boneanomaly(276 boxes),bonelesion(45 boxes),foreignbody(8 boxes),fracture(18090 boxes),metal(818 boxes),periostealreaction(3453 boxes),pronatorsign(567 boxes),softtissue(464 boxes),text(23722 boxes).
Fig1: few random testing images visulaization within their ground-truth label and relative bounding box.
The data are already annotate in many different formats, one of the is the YOLO one. Indeed, YOLOv7 is the chosen model for this project.
First download the YOLOv7 repo:
git clone https://github.com/WongKinYiu/yolov7.git
cd yolov7
Then download the GRAZPEDWRI-DX dataset and stored it in the GRAZPEDWRI-DX_dataset folder. Keep the YOLO annotations (.txt files) and extract the images from the .zip into the images folder.
conda create -n yolov7 python=3.9
conda activate yolov7
pip install -r requirements.txt
The dataset iamges and annotatation are contained into data/full/images and data/full/labels respectively. The script:
python split.py
will divide the dataset 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
├── yolov5
│ ├── images
│ └── labels
├── 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.
It is available the ONNX model in runs/train/yolov7/weights/yolov7-p6-bonefracture.onnx, and inferece can be performed:
python inference_onnx.py --model-path <onnx model path> --img-path <input image path> --dst-path <destination folder for predictions>
for example:
python inference_onnx.py --model-path runs/train/yolov7/weights/yolov7-p6-bonefracture.onnx --img-path ./GRAZPEDWRI-DX_dataset/images/test/0038_0775938745_03_WRI-L2_M014.png --dst-path ./predictions
The results (the predicted labelled image in .png file, and the predicted labels in .txt file) in the destination folder indicated. Each detection is saved as:
<class> <confidence_score> <bbox_Xcenter> <bbox_Ycenter> <bbox_Width> <bbox_Height>
N.B.: remeber to install the requirements:
pip install onnx onnxruntime
python app/app.py
GUI icons were made by Freepik, smashingstocks, photo3idea_studio, Yogi Aprelliyanto, Muhammad Ali, and dmitri13 from www.flaticon.com.
The GUI is done using PySide6, so first install it with:
pip install PySide6
The torch model is also available in runs/train/yolov7/weights/yolov7-p6-bonefracture.pt and you can evaluate it or perform inference using test.py and detect.py script respectively of the YOLOv7 repo. On the contrary, if you want to train the model, use train.py.
The evaluation of the 3 different datasets is collected in the runs/test folder. For each dataset are saved the predicted labels by the model, the confusion matrix, the F1, P, R, and PR curve plot.
You can use data_overview.ipnb to describe a dataset (.csv file) and visulize the ground-truth labels with or without the predicted ones as in Figure 2 here below.
Fig2: few random testing images visualization within their realive ground-truth and the predicted labels with their confidence score by the trained YOLOv7-p6 model.