Combining medical image datasets improves some lesion detection tests
Researchers combined datasets with different labels; detection gains varied by finding in the reported tests.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Sheethal Bhat, Bogdan Georgescu, Awais Mansoor, Florin C. Ghesu, Sasa Grbic, Andreas Maier
- Universities and research institutions
- Friedrich-Alexander-Universität, 91058, Erlangen-Nürnberg, Germany, Siemens Healthineers, 91052, Erlangen, Germany, Siemens Medical Solutions, 08540, Princeton, USA
Institution metadata: Original paper ↗
What they did and found
Researchers trained separate models on four medical image datasets, used them to fill missing labels, and trained a combined detector. Compared with a single-dataset detector, the new model improved detection on several chest X-ray and mammography findings. Gains were limited for atelectasis.
Why it matters
The tests show a way to combine datasets with different labels, but they do not establish how the model would perform in clinical use.
Separate models predicted missing findings in the other datasets. The final model learned from both expert annotations and these generated labels, aiming to retain what each dataset contributed.
The combined model improved detection for several findings in the reported tests. Atelectasis showed smaller gains because its labels can overlap with other conditions and vary across datasets.
What remains uncertain
The study tested selected chest X-ray and mammography datasets. Improvements varied by finding, and the paper does not report a clinical deployment.
Original sources · 1
Check the original paper for its authors, methods, version and access terms.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.