The CAS2023 dataset is an MR modality segmentation dataset featured in this year's MICCAI 2023 Challenge. The officials provided a training set of 100 cases, with only cerebral arteries annotated. The context of the dataset is cerebrovascular diseases, which are one of the highest concerns in brain health, with both the cerebral vessels themselves and the brain hemorrhages from strokes being fine and complex structures that are difficult to accurately segment by current artificial intelligence systems and even by doctors. This dataset is obtained via Magnetic Resonance Angiography (MRA). MRA is commonly used to display the cerebral arterial tree to assist in diagnosis. Accurate segmentation of MRA cerebral arteries is crucial for quantitative analysis, such as estimating the degree of luminal stenosis. However, manual segmentation is challenging for experts due to the complexity of the cerebral artery network, individual variability, and weak signals from small vessels. Time-of-Flight (TOF) MRA is a widely used non-invasive imaging technique that can delineate the cerebral vasculature tree without the need for contrast agents. However, the lack of large-scale, open-access TOF-MRA data with well-labeled cerebral arteries limits the development and validation of reliable automatic cerebral artery segmentation algorithms. Therefore, validating cerebral vascular segmentation algorithms on TOF-MR has certain clinical significance.
| Dimensions | Modality | Task Type | Anatomical Structures | Anatomical Area | Number of Categories | Data Volume | File Format |
|---|---|---|---|---|---|---|---|
| 3D | TOF-MR | Segmentation | cerebral artery | Brain | 1 | 100 | .nii.gz |
| Dataset Statistics | spacing (mm) | size |
|---|---|---|
| min | (1.00, 1.00, 1.00) | (208, 320, 96) |
| median | (1.00, 1.00, 1.00) | (640, 640, 150) |
| max | (1.00, 1.00, 1.00) | (784, 784, 255) |
Number of two-dimensional slices in the dataset: 14,349 (based on statistics from 100 training cases).
| Metric | Tumor |
|---|---|
| Case Count | 100 |
| Coverage | 100% |
| Min Volume (cm³) | 17 |
| Median Volume (cm³) | 170 |
| Max Volume (cm³) | 331 |
It follows the default nnUNet format:
Dataset
│
├── imagesTr
│ └── ...
├── labelsTr
│ └── ...
├── dataset.json
Huijun Chen (Tsinghua University)
Xihai Zhao (Tsinghua University)
Haozhong Sun (Tsinghua University)
Jiaqi Dou (Tsinghua University)
Chenlin Du (Tsinghua University)
Runyu Yang (Tsinghua University)
Xiaoqi Lin (Tsinghua University)
Han Jiang (Tsinghua University)
Shuwan Yu (Tsinghua University)
Jiachen Liu (Tsinghua University)
Zixuan Han (Tsinghua University)
Chun Yuan (University of Washington)
Niranjan Balu (University of Washington)s
Official Website: https://codalab.lisn.upsaclay.fr/competitions/9804
Download Link: https://codalab.lisn.upsaclay.fr/competitions/9804#participate-get_starting_kit
Article Address: TBD
Publication Date: April, 2023.
TBD
Original introduction article is here.