Adapting DINOv2 to 3D and pretraining on ~100,000 unlabeled MRI, CT, and PET scans yields a general 3D medical imaging ViT that improves downstream segmentation and classification, especially with little labeled data.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.IV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A generalizable 3D framework and model for self-supervised learning in medical imaging
Adapting DINOv2 to 3D and pretraining on ~100,000 unlabeled MRI, CT, and PET scans yields a general 3D medical imaging ViT that improves downstream segmentation and classification, especially with little labeled data.