Pyramid Self-Contrastive Learning for Single-shot Test-time Ultrasound Image Denoising
Researchers have developed a novel Pyramid Self-Contrastive Learning (PSCL) framework designed to denoise ultrasound images without requiring any prior training. This method operates on single-shot imaging, utilizing multiple noisy samples to disentangle anatomical similarity from noise within separate pyramid latent spaces. The framework then reconstructs a clean image by isolating the anatomical information. Experiments on synthetic aperture ultrasound (SAU) demonstrated significant improvements in signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR), with real-world applications on cardiac, liver, and kidney imaging showing substantial gains in image clarity. AI
IMPACT This novel approach could lead to clearer medical imaging without the need for extensive training data or domain-specific models.