Medical Imaging · wanglab
CT_DeepLesion-MedSAM2
CT lesion segmentation mask annotations generated using MedSAM2 for model training and evaluation.
What CT_DeepLesion-MedSAM2 is
CT_DeepLesion-MedSAM2 is a medical imaging dataset providing segmentation mask annotations for computed tomography (CT) scans derived from the DeepLesion benchmark. Developed by researchers at institutions including University Health Network, Vector Institute, and Harvard Medical School, it restructures lesion records into segmentation masks suitable for 3D medical foundation models.
The underlying DeepLesion collection contains 32,735 diverse lesions across 32,120 CT slices from 10,594 studies of 4,427 unique patients. CT_DeepLesion-MedSAM2 adapts a subset of these cases to facilitate the training and evaluation of MedSAM2, a foundation model designed to segment structures across 3D medical scans and video sequences.
How it was made
In the original DeepLesion dataset, each lesion has a bounding box annotation on a key CT slice determined from its longest diameter and perpendicular diameter. For this dataset, 5,000 of these lesions were segmented using MedSAM2 within a human-in-the-loop pipeline to produce detailed lesion masks.
Key facts
- Modality
- Computed tomography (CT)
- Annotated samples
- 5,000 lesions
- Source repository
- DeepLesion
- Annotation method
- MedSAM2 human-in-the-loop pipeline
- Data formats
- CSV, tabular
What you can use it for
- Segmentation model training
- Training MedSAM2 and other medical image segmentation foundation models on 3D CT lesion boundaries.
- Benchmark evaluation
- Assessing automated CT lesion segmentation algorithms against human-in-the-loop mask annotations.
- Universal lesion detection research
- Developing algorithms that identify and delineate diverse lesion types across varied anatomical regions.
Explore the data
No licence declared
The authors have not published a licence, which means no permission to redistribute the data has been granted to anyone. The rows are not served here; read them on the original dataset page.Strengths and limits
Good at
- Converts original diameter-derived bounding boxes into detailed segmentation masks for 5,000 lesions.
- Uses a human-in-the-loop pipeline to verify and refine model-assisted segmentations.
- Leverages diverse clinical cases from the multi-patient DeepLesion benchmark.
Watch out for
- Mask annotations cover only 5,000 of the 32,735 lesions catalogued in the full DeepLesion dataset.
- The dataset is restricted entirely to computed tomography and does not include other imaging modalities.
Questions
Can I look inside CT_DeepLesion-MedSAM2 without downloading it?
Not here: it declares no licence, so no permission to redistribute its rows has been granted. Its description and statistics are still shown.
What licence is CT_DeepLesion-MedSAM2 under?
Its authors have not declared a licence, which means no permission to reuse or redistribute it has been granted.
What is the relationship between this dataset and DeepLesion?
DeepLesion provides bounding boxes based on clinical diameter measurements across 32,735 lesions. This dataset provides segmentation masks for 5,000 of those lesions using MedSAM2.
How were the segmentation masks generated?
Masks were generated using the MedSAM2 foundation model within a human-in-the-loop pipeline.
What imaging modality is included?
The dataset contains computed tomography (CT) image references and annotations.
How to cite it
Ma, J., Yang, Z., Kim, S., Chen, B., Baharoon, M., Fallahpour, A., Asakereh, R., Lyu, H., & Wang, B. (2025). MedSAM2: Segment Anything in 3D Medical Images and Videos. arXiv preprint arXiv:2504.63609.
Written from CT_DeepLesion-MedSAM2's own dataset card. Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.