A Dual-Stage Denoising Method Based on Zero-Shot Learning for Low-Field MRI

Motivation: Low-field MRI is constrained by the physical factors of detection equipment, facing issues such as noise that degrade image quality and affect disease diagnosis. Goal(s): This study aims to denoise low-field magnetic resonance images using deep learning techniques. Approach: This paper proposes a dual-stage denoising method based on zero-shot learning: the first stage uses a supervised deep learning method for denoising, while the second stage employs a zero-shot denoising method. Results: Results demonstrate that the dual-stage denoising method outperforms both the supervised method and the zero-shot denoising method when applied individually, effectively achieving an improvement in the quality of low-field magnetic resonance images. Impact: The framework of our proposed dual-stage denoising method is plug-and-play for various existing denoising models and generally enhances their performance.