336,000 chest X-ray images are labeled with 20 harmonized clinical findings. The dataset includes uncertainty flags for annotations and provides patient-level splits for training and evaluation.
Use Cases
- Train multi-label classification models based on the 20 harmonized clinical labels.
- Develop uncertainty-aware diagnostic algorithms based on the provided uncertainty flags.
- Benchmark model performance on patient-level splits to avoid data leakage.
- Study label consistency and harmonization across different radiology datasets.
Strengths
- 336,000 chest X-ray images provide a substantial sample size.
- 20 harmonized labels offer a standardized set of clinical findings.
- Patient-level splits help ensure proper evaluation of model generalization.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Description metadata is limited; actual data quality requires manual inspection after download.