You can find the corresponding journal articel at https://www.urncst.com/index.php/urncst/article/view/789

To generate a random selection of size 224x224 images:

1. Download original dataset from https://data.mendeley.com/datasets/56rmx5bjcr/1 Complete the following steps within the prepareKneeOsteoarthritisXray folder
2. Create folder named all_data
3. Create a template_empty_folder with two subfolders train and test and each of those with two subfolders 0 and 1
4. Copy and paste template_empty_folder twice and rename to selected_data and selected_processed_data
5. Move contents of KneeOsteoarthritisRNN/ClsKLData/kneeKL224 into all_data (excluding "val" and auto_test folders)
6. Run the python scripts in terminal
   • cd my/path/to/prepareKneeOsteoarthritisXray
   • python ./datasplitting.py
   • python ./preprocessing.py
7. (Optional) manually delete images with artifacts like implants
8. (Optional) color invert images that are randomly color inverted the wrong way

To generate a random selection of size 384x384 images:

1. Follow the above steps, but in addition:
   • Uncomment line 35 under "RESIZING" in preprocessing.py
   • Move contents of KneeOsteoarthritisRNN/ClsKLData/kneeKL299 instead of kneeKL224 (again excluding "val" and "auto_test" folders)

To redo selection process:

1. Replace selected_data and selected_processed_data with template_empty_folder and rename both again
2. Repeat remaining steps

Note: ratios are off slightly due to "0" category being taken from two folders and "1" category from three folders

• train/0: 500
• train/1: 498
• test/0: 100
• test/1: 99

You are now ready to fine-tune using the fine-tune.py script