Automated Periodontitis Bone Loss Diagnosis in Panoramic Radiographs Using a Bespoke Two-Stage Detector
All data is hosted on Google Drive:
| Path | Size | Files | Format | Description |
|---|---|---|---|---|
| perio-dataset | ---- | ---- | Main folder | |
| via_export_coco_BL.json | 14.3 MB | 1 | JSON | Metadata, COCO format annotations for bone loss. |
| via_export_coco_FI.json | 14.3 MB | 1 | JSON | Metadata, COCO format annotations for furcation involvement. |
| Images.zip | 2.72 GB | 1 | ZIP | Panoramic radiographs images at 1536x2976, ZIP archive. |
This network is use for automatic periodontitis bone loss diagnosis, if more background information is needed please refer to the paper.
This code is written in python language with Keras framework. This code contains the network training part and the prediction part, and the experiment model weights and datasets can be provided if needed.
1. Preparation of dataset
train.txt and val.txt contain the path and annotation information of
the image train and validation samples.
Keep your images in a fold, and create two txt documents
to write information in the following form.
xxx/Images/1.jpg 810,620,957,901,0 944,617,1106,916,1 ...
xxx/Images/2.jpg 810,620,957,901,1 944,617,1106,916,0 ...
...
2. Train
After preparing the dataset, complete the python environment, path information, hyperparameters and other settings, you can start training with train.py. You can save intermediate models on demand.
3. Predict
You can generate a display of the model's predictive effect on the test sample with predict.py and to ensure the script runs correctly please check the path dependencies.
https://github.com/qqwweee/keras-yolo3/
https://github.com/jinfagang/keras_frcnn
https://github.com/Cartucho/mAP
https://github.com/bubbliiiing/centernet-keras