数据集获取

git clone https://github.com/SCSE-Biomedical-Computing-Group/CT-Subcortical-Segmentation.git

curl -L -o repo.zip https://github.com/SCSE-Biomedical-Computing-Group/CT-Subcortical-Segmentation/archive/refs/heads/main.zip
unzip repo.zip

源站 README 摘录（使用方式）

CT Brain Subcortical Segmentation

The human subcortex contains many structures that hold key responsibilities in many physiological functions fundamental to daily living. These structures have been found to exhibit volume and shape changes during development of neurodegenerative disorders and hence, automated segmentation of these subcortical anatomies will be meaningful for future studies on understanding and developing computer-aided solutions for neurological conditions. The main contributions of this research are:

• Proposal of an automated cross-domain subcortical segmentation label generation framework
• Generation of the first publicly available CT subcortical segmentation dataset
• Training and comparison of various established and SOTA models for CT subcortical segmentation

Repository Organisation

The code and outputs of this research are organised according to the main contributions in three main folders.
label_generation_framework contains the relevant code for the label generation framework as well as that for consensus voting.
subcortical_segmentation_dataset contains the generated subcortical segmentation masks. The generated subcortical masks contains 17 subcortical regions. The label mapping is available in label_mapping.txt.
transfer_learning contains the notebooks and model weights used for the transfer learning experiments to validate the utility of our generated CT subcortical segmentation dataset.
segmentation_models_training contains the script for running the training of the segmentation models. The models trained include 2D UNet, 3D UNet, Swin UNETR and nnU-Net.

Set-Up

To run the code in this repository, it is recommended to create a new python/conda virtual environment and install the packages in the requirements.txt file.

Automated Cross-Domain Label Generation Framework

• framework.py contains the code necessary to run the both ensembling methods (consensus and majority voting) on the models’’ outputs to generate new masks.

Generated Subcortical Segmentation Dataset

• The dataset only contains the generated masks and not the actual CT scans as we do not own the CT scans data and they should be obtained the SynthRAD 2023 Challenge (Task 1) page itself.

• Upon obtaining the CT scans, the scans should be processed using the mri_convert tool in the FreeSurfer package. The steps are as follo

数据加载示例（图像类）

from PIL import Image
import glob, os

files = (glob.glob(os.path.join(path, "**", "*.png"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.jpg"), recursive=True)
       + glob.glob(os.path.join(path, "**", "*.tif"), recursive=True))
print("图像文件数:", len(files))
img = Image.open(files[0]); print("尺寸/模式:", img.size, img.mode)

# torchvision Dataset 方式:
# from torchvision import datasets
# ds = datasets.ImageFolder(path)  # 要求 子目录=类别

目录组织与标注格式以源站说明和下载后实际文件为准。

完整仓库：github.com/SCSE-Biomedical-Computing-Group/CT-Subcortical-Segmentation