Author: Burak Demir, MD, FEBNM
Version: 1.0
Contact: 4burakfe@gmail.com
SlicerPETDenoise is a 3D Slicer extension that includes three modules intended for medical image research, particularly PET and CT image workflows. These modules assist in denoising, comparing, and visually fusing volumetric medical data. The tools are developed with research utility in mind and are not intended for clinical use.
Related work: Demir, B., Atalay, M., Yurtcu, H. et al. Denoising of PET with SwinUNETR neural networks: impact of tumor oriented loss function, denoising module for 3D slicer. Ann Nucl Med (2026). https://doi.org/10.1007/s12149-026-02166-4
This extension is not currently on Extension Manager of the 3D Slicer but can be installed manually. For installation download this repository and extract the zip folder. Then in the 3D Slicer go to Edit->Application Settings->Modules->Additional Module Paths Here click >> button and manually add PETDenoise, VolumeComparator and Easy_fusion folders.
SlicerPETDenoise, SlicerVolumeComparator requires Pytorch to be installed to operate. You can install it from extension manager with PyTorchUtils extension. If Pytorch Utils is not installed the modules will not be shown.
It is highly recommended to have CUDA capable GPU and if so be sure you have installed CUDA enabled version of Pytorch. Otherwise, it will work very very slow.
You can train your own models with scripts provided here: https://github.com/4burakfe/Claritas
This repository also contains pretrained models ready for use: https://github.com/4burakfe/SlicerPETDenoise/releases/tag/Models
You can test this module with the cases in here: https://github.com/4burakfe/SlicerPETDenoise_SampleCases/releases/tag/images
Performs AI-based denoising of PET volumes using deep learning models. Supports UNET, SwinUNETR, and SwinUNETR+GCFN architectures.
- Accepts PET alone or PET+CT (dual channel)
- Loads
.pthmodels with associated.txtconfiguration - Applies sliding window inference on volumes
- CUDA acceleration (if available)
- Select input PET volume.
- (Optional) Select corresponding CT volume for dual-channel denoising.
- Choose model architecture and parameters (UNET/SwinUNETR).
- Select a folder containing
.pthmodel and config.txtfile. - Adjust voxel spacing and block size if needed.
- Click Denoise. The output will be a new volume named like
OriginalName_DN_ModelName.
Notes:
- You may be prompted to install required Python packages (
monai,einops).- Ensure the model folder and config file are correctly formatted.
Computes numerical differences between two volumes using several image similarity metrics.
- Mean Squared Error (MSE)
- Mean Absolute Error (L1)
- Structural Similarity Index (SSIM)
- PSNR
- Edge Loss (via gradient differences)
- Load two volumetric datasets (same shape and resolution).
- Enter SSIM dynamic range value (default: 10).
- Click Compare.
- Log area will display computed metrics.
Note: PyTorch and MONAI are required.
Provides basic image fusion for PET/SPECT and CT/MRI volumes. Includes tools for PET color mapping, CT windowing, and 3D view control.
- Load both SPECT/PET and CT/MR volumes.
- Use buttons to select:
- PET color map (Inferno, Hot Iron, etc.)
- CT windowing (head, abdomen, etc.)
- View rotation and orientation
- Click Go to initiate 3D volume rendering and MIP setup.
- Use toggle to start/stop rotation or change angle.
Note: This module only affects visualization, not image content.
- Install 3D Slicer
- Clone or download this repository.
- This extension is not currently on Extension Manager of the 3D Slicer but can be installed manually. For installation download this repository and extract the zip folder. Then in the 3D Slicer go to Edit->Application Settings->Modules->Additional Module Paths Here click >> button and manually add PETDenoise, VolumeComparator and Easy_fusion folders.
This extension uses several Python libraries inside Slicer's environment:
torchmonaieinops
Torch must be installed with PyTorchUtils extension... If PyTorchUtils is not installed the modules will not appear.