DuDoDp-MAR

This repository contains the official implementation of the paper "Unsupervised CT Metal Artifact Reduction by Plugging Diffusion Priors in Dual Domains" by [Xuan Liu et al.]. The paper introduces an unsupervised method for metal artifact reduction using diffusion priors.

Test Environment

OS: Ubuntu 20.04
GPU: NVIDIA RTX 3090
Python (=3.9)
Pytorch (=1.13.1)
Torchvision (=0.14.1)

Requirements

pip install -r requirements.txt

Pre-trained Models

Download from Google Drive.

Test data

Please refer to SynDeepLesion.

How to run

python mar.py -c config/MAR.yaml

Important parameters in configuration yaml file:

# training data
...
# model
...
# diffusion
...
# train
...
# sample
model_path: 'Patch-diffusion-pretrained/model150000.pt' # Path of pre-trained model$
...
timestep_respacing: [10, 10, 10, 10, 10, 10, 10, 10, 10, 10] # Acceleration
# MAR
a: 0.4 # \delta(t) = (a-1)e^{-n\frac{t}{T}}+1
n: 4 # \delta(t) = (a-1)e^{-n\frac{t}{T}}+1
delta_y: 0.8 # \mathcal{M}_y = \mathcal{M}(\delta_y).
save_dir: 'results/DuDoDp-MAR' # Path of MAR results
data_path: './test_data/SynDeepLesion' # Path of test data
inner_dir: 'test_640geo/' # be 'apdcephfs/share_1290796/hazelhwang/mardataset/test_640geo/' for data from https://github.com/hongwang01/SynDeepLesion
# root 
# ├── test_640geo_dir.txt 
# ├── testmask.npy
# └── inner_dir
#          ├── ...
#          └── patient
#               ├── ...
#               └── slice
#                     ├── 0.h5
#                     ├── ...
#                     └── gt.h5
num_test_image: 1 # num of test image, 200 for all test images
num_test_mask: 10 # number of test masks

Acknowledgment

Big thanks to PatchDiffusion-Pytorch and guided-diffusion for providing the codes that facilitated the training of diffusion models, and SynDeepLesion for the test data.