Pyramid Self-Contrastive Learning Framework for Test-time Ultrasound Image Denoising
A new test-time training framework for one-shot ultrasound image denoising is proposed, specifically applied to synthetic aperture ultrasound (SAU). The Aperture-to-Aperture (A2A) framework uses self-contrastive learning in pyramid latent spaces to separate anatomical similarity from noise randomness in shuffled sub-apertures. The clean image is then decoded from the anatomy space. This approach addresses domain shift issues in complex in vivo environments, avoiding reliance on explicit noise assumptions or massive labeled data. The method is detailed in arXiv:2605.12567v1.
Key facts
- Proposes a pure test-time training framework for one-shot ultrasound image denoising.
- Applied to synthetic aperture ultrasound (SAU).
- Aperture-to-Aperture (A2A) framework disentangles anatomical similarity and noise randomness.
- Uses self-contrastive learning in pyramid latent spaces.
- Clean image is decoded from the anatomy space.
- Addresses domain shift in complex in vivo environments.
- Avoids reliance on explicit noise assumptions or massive labeled data.
- Published on arXiv with ID 2605.12567v1.
Entities
Institutions
- arXiv