Our model, trained on clinically representative animal dataset, significantly improves preservation of small anatomical structures over models trained on human dataset.

It improves image quality, without hallucinations or loss of detail.