zinhoo / dudonet

This repo is used to prepare metal artifact dataset in DuDoNet and DuDoNet++. The clean and metal contaminated CT images are illustrated below, respectively. Note the red pixels indicate the metallic implants.

1. Create folders database_MAR/images_train and database_MAR/images_val.

2. We put a sample image images/000001_01_01 to run the dome prepare_image_database.m.

3. To simulate the whole MAR database. Download CT images according to step 3. Uncomment line 120 and comment ling 121 in prepare_image_database.m. Mode train\val can be selected in line 32. Note: mode = 'train' will take more than 5 hours and require ~560 G hard disk storage.

4. Download and unzip DeepLesion dataset to folder images. Images_png_01.zip~Images_png_10.zip is sufficient. The structure of the unzipped files should look like:

   images/
       000001_01_01/
       000001_02_01/
       ...
   

Thanks Dr. Yanbo Zhang and Dr. Hengyong Yu for providing the metal artifact synthesis code.