The code for the paper "Hybrid CNN-Transformer Network with Circular Feature Interaction for Acute Ischemic Stroke Lesion Segmentation on Non-contrast CT Scans" submitted to IEEE TMI.
CUDA 11.4
Python 3.8.12
Pytorch 1.11.0
Torchvision 0.12.0
batchgenerators 0.21
SimpleITK 2.1.1
scipy 1.8.0
- Install nnUNet and ClSeg as below
cd nnUNet
pip install -e .
cd ClSeg_package
pip install -e .
AISD dataset can be downloaded from (https://github.com/griffinliang/aisd). Pre-process the datasets using the preprocess codes in nnUNet/nnunet/dataset_conversion.
all compared methods use the same pre-processing steps as nnUNet. The documentation of the pre-processing can be found at [DOC]
cd ClSeg_package/ClSeg/run
- Run
python run_training.py -network_trainer nnUNetTrainerV2_AISD -gpu='0' -task={task_id} -outpath='AISD'
for training.
- Run
python run_training.py -network_trainer nnUNetTrainerV2_AISD -gpu='0' -task={task_id} -outpath='AISD' -val --val_folder='validation_output'
for testing.
The pre-trained model of AISD dataset can be downloaded from [Baidu YUN] with the password "4phx".
During reproduction, for the methods (e.g. LambdaUNet [1], UNet-AM [2], UNet-GC [3]) that do not publish their codes, we endeavored to implement their approaches by following the technical details provided in their papers. our reproduced codes of these methods can be found at [DOC] and [DOC]
For the compared methods with Open-source codes, we directly use their codes for AIS lesion segmentation on 2 AIS datasets. The links of their open-source codes can are listed as follows:
[AttnUnet2D]
[Swin-Unet]
[TransUNet]
[FAT-Net]
[AttnUNet3D]
[nnFormer]
[UNETR]
[CoTr]
[nnUNet]
[UNet-RF]
Note that for all compared methods, to perform fair comparisons, we use the same pre-processing steps (as 1.2 Pre-prcoessing) and the same data split. and in the paper all compared results were derived from our reproduction experiments.
Part of codes are reused from the nnU-Net. Thanks to Fabian Isensee for the codes of nnU-Net.