---

DenoMamba
_{A fused state-space model for low-dose CT denoising}

Şaban Öztürk; Oğuz Can Duran; Tolga Çukur;

---

[arXiv]

Official PyTorch implementation of DenoMamba, a novel denoising method based on state-space modeling (SSM), that efficiently captures short- and long-range context in medical images. Following an hourglass architecture with encoder-decoder stages, DenoMamba employs a spatial SSM module to encode spatial context and a novel channel SSM module equipped with a secondary gated convolution network to encode latent features of channel context at each stage. Feature maps from the two modules are then consolidated with low-level input features via a convolution fusion module (CFM).

Dependencies

python>=3.8.13
cuda=>11.6
torch>=2.2
torchvision>=0.17
pillow
scipy
mamba-ssm==1.1.3

You are welcome to ask any library issues.

Dataset

To reproduce the results reported in the paper, we recommend the following dataset processing steps:

Sequentially select subjects from the dataset. Select 2D cross-sections from each subject. Normalize the selected 2D cross-sections before training and before metric calculation.

You should structure your aligned dataset in the following way:

/datasets/LDCT/
  ├── Full_Dose                 ├── Quarter_Dose
  │   ├── 1mm                   │   ├── 1mm
  │     ├── Sharp Kernel        │     ├── Sharp Kernel
  │       ├── L506              │       ├── L506
  |          - ...              |          - ...
  │       ├── ...               │       ├── ...
  │     ├── Soft Kernel         │     ├── Soft Kernel
  │       ├── L506              │       ├── L506
  |          - ...              |          - ...
  │       ├── ...               │       ├── ...
  │   ├── 3mm                   │   ├── 3mm
  │     ├── Sharp Kernel        │     ├── Sharp Kernel
  │       ├── L192              │       ├── L192
  |          - ...              |          - ...
  │       ├── ...               │       ├── ...
  │     ├── Soft Kernel         │     ├── Soft Kernel
  │       ├── L192              │       ├── L192
  |          - ...              |          - ...
  │       ├── ...               │       ├── ... 

Training

Example Command for training with default arguments

python3 train.py --full_dose_path example_full_dose_path --quarter_dose_path example_quarter_dose_path --path_to_save example_path_to_save_the_trained_model

Argument Descriptions

Argument	Description
--full_dose_path	Path to full dose images
--quarter_dose_path	Path to quarter dose images
--dataset_ratio	The ratio of dataset to use (in case of big dataset)
--train_ratio	Ratio of train dataset to all dataset
--batch_size	Batch size.
--in_ch	Number of input image channels
--out_ch	Number of output image channels
--learning_rate	Learning rate
--max_epoch	Maximum number of epochs
--continue_to_train	Continue any interrupted training
--path_to_save	Path to save the trained model
--ckpt_path	Path to trained and saved checkpoint model
--validation_freq	Frequency to run validation
--save_freq	Frequency to save model
--batch_number	Number of a batch in validation to show the sample images
--num_blocks	Number of FuseSSM block layers
--dim	Number of FuseSSM blocks
--num_refinement_blocks	Number of refinement blocks

Key Required Arguments

--full_dose_path: Must specify the path to the full-dose dataset.
--quarter_dose_path: Must provide the path to the quarter-dose dataset.
--path_to_save: Required to specify where the trained model will be saved.

These arguments are essential for the program to locate the necessary datasets and save the trained model.

Test

Example Command for test with default arguments

python test.py --full_dose_path example_full_dose_path --quarter_dose_path example_quarter_dose_path --path_to_save example_path_to_save_the_trained_model --output_root example_path_to_save_the_output_images

Argument Descriptions

Argument	Description
--full_dose_path	Path to full dose images
--quarter_dose_path	Path to quarter dose images
--dataset_ratio	The ratio of dataset to use (in case of big dataset)
--train_ratio	Ratio of train dataset to all dataset
--batch_size	Batch size.
--in_ch	Number of input image channels
--out_ch	Number of output image channels
--ckpt_path	Path to trained and saved checkpoint model
--output_root	Path to save denoised images
--num_blocks	Number of FuseSSM block layers
--dim	Number of FuseSSM blocks
--num_refinement_blocks	Number of FuseSSM blocks

Key Required Arguments

--full_dose_path: Must specify the path to the full-dose dataset.
--quarter_dose_path: Must provide the path to the quarter-dose dataset.
--ckpt_path: Required to specify where the trained model was saved.
--output_root : Required to specify where to save the output images.

These arguments are essential for the program to locate the necessary datasets and save the trained model.

Citation

You are encouraged to modify/distribute this code. However, please acknowledge this code and cite the paper appropriately.

@article{ozturk2024denomamba,
  title={DenoMamba: A fused state-space model for low-dose CT denoising}, 
  author={Şaban Öztürk and Oğuz Can Duran and Tolga Çukur},
  year={2024},
  journal={arXiv:2409.13094}
}

For any questions, comments and contributions, please feel free to contact Şaban Öztürk (saban.ozturk[at]bilkent.edu.tr)

Acknowledgments

This code uses libraries from Restormer and mamba repository.

---

Copyright © 2024, ICON Lab.