This repsotiry is about lesion segmentation on multi modality images including CT, PET/CT, and SPECT/CT. All models have been trained using nnU-Net. You can find inference instructions in the provided here file. It includes easy-to-use code for handling large images and several useful options for efficient inference on large datasets. You can also select the appropriate input image modality based on your needs. Currently, the code supports only the NIfTI file format, so you’ll need to convert your DICOM images to NIfTI before running the code. Instructions and scripts for converting DICOM to NIfTI will be available shortly. Alternatively, you can follow the standard nnU-Net inference instructions available on nnU-Net after downloading the trained models folder.
This is the updated models trained on dataset provided by by Jafari et al published in EJNMMI available here. The models require body-weight SUV unit and HU CT images as input. Separate models are available using PET only, CT only, and both PET and CT (PET/CT) images. Please ensure you select the appropriate model based on your input. We recommend using the PET only segmentation model when there is a respiratory misalignment between PET and CT images to ave a more accurate output. for the cases with perfect alignment we recommend using PET/CT models. The presence of respiratory misalignment can be automatedly checked using this resposiroy. there are two versions of trained models, first using the old nnU-Net network architecture, and the second using the new nnU-Net architceture using residual blocks. the comparison can be found here
This nnUNet models are trained on a big multi-centric local and public dataset including various malignancies.
Before running the tools, make sure your GPU drivers are properly installed and that nnunetv2 is installed as per their instructions, which can be found at: https://github.com/MIC-DKFZ/nnUNet. Alternatively, you can install nnunetv2 using the following commands:
pip install --upgrade git+https://github.com/MIC-DKFZ/nnUNet.git
pip install --upgrade git+https://github.com/FabianIsensee/hiddenlayer.gitTo install the organ segmentation repository and use the manual inference instruction, simply run:
pip install git+https://github.com/YazdanSalimi/Organ-Segmentation.gitWe welcome any feedback, suggestions, or contributions to improve this project!
for any furtehr question please email me at: [email protected]