Pre-extracted embeddings from the MIMIC-CXR chest X-ray database using five vision transformer architectures. The dataset is organized by 20 seeds and 3 coreset selection strategies per seed, totaling approximately 1,999 to 2,372 samples. It was created by MITCriticalData and last updated on March 7, 2026.
Use Cases
- Training quantum machine learning models based on pre-computed vision transformer embeddings.
- Benchmarking coreset selection strategies for efficient model training based on the three strategies per seed.
- Exploring transfer learning for medical imaging tasks based on embeddings from five state-of-the-art vision models.
Strengths
- Embeddings derived from the established MIMIC-CXR database.
- Includes outputs from five different vision transformer architectures.
- Organized with 20 seeds and 3 coreset strategies per seed for experimental rigor.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- MIMIC-CXR Database
- Collection Method
- Pre-extracted embeddings from chest X-ray images using multiple vision models.
- Time Range
- null
- Freshness
- Last updated 2026-03-07 13:40:59; freshness should be verified.
- Geography
- null