Combining contrast-aware data augmentation with domain-adversarial training helps deep learning MRI reconstruction models generalize from adult to neonatal data, achieving better image quality when scanning with limited data.
This paper improves how AI models trained on adult brain MRI scans work on neonatal (newborn) brain scans. The researchers used two techniques: augmenting training data to mimic neonatal contrast differences, and adversarial training to reduce domain shift. Results show these methods significantly improve reconstruction quality at high acceleration factors.