A large-scale dataset constructed for medical visual question answering (Med-VQA) tasks. It is based on the ReXGroundingCT data and contains CT volumes paired with multi-class segmentation masks, where each mask channel represents a specific lesion type. The dataset was uploaded by liyf001 and last updated on October 11, 2025.
Use Cases
- Train Med-VQA models based on CT volumes and corresponding questions.
- Evaluate model performance on lesion identification tasks based on multi-class segmentation masks.
- Develop segmentation-aided VQA systems based on the alignment of lesion masks and CT volumes.
Strengths
- Large-scale size is explicitly mentioned in the description.
- Includes multi-class segmentation masks for specific lesion types.
Limitations
- Row count, column names, and file formats are unknown, limiting suitability assessment.
- Column-level documentation is absent; field semantics must be inferred after download.
Provenance
- Source
- Based on the ReXGroundingCT data.
- Collection Method
- Constructed for model training and evaluation.
- Freshness
- Last updated 2025-10-11 10:14:04.