Low-field MRI image enhancement method based on adaptive noise perception(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2026年第3期
- Page:
- 293-299
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Low-field MRI image enhancement method based on adaptive noise perception
- Author(s):
- LIU Qi1; HU Yutao2; YANG Chunfeng2; CHEN Yang1; 2
- 1. School of Biomedical Engineering, Southern Medical University, Guangzhou 510515, China 2. School of Computer Science and Engineering, Southeast University, Nanjing 211189, China
- Keywords:
- Keywords:?ow-field magnetic resonance imaging image enhancement deep learning dual-path network
- PACS:
- R318；R445.2
- DOI:
- DOI:10.3969/j.issn.1005-202X.2026.03.003
- Abstract:
- Abstract: Objective To achieve efficient and robust quality enhancement for low signal-to-noise ratio (SNR) and high-noise-level magnetic resonance imaging (MRI) images acquired by low-field MRI devices. Methods Given the lack of one-to-one paired high- and low-field MRI images, a simulation method was develop to generate low-field MR images by adding noise into the K-space data of high-field MRI images and performing undersampling in the image domain, thereby constructing a paired high- and low-field MRI image dataset for training. A dual-path denoising and enhancement network with an adaptive noise perception branch was introduced to tackle the difficulty in perceiving noise intensity in low-field MR images. Publicly available datasets IXI and fastMRI were used to establish training datasets for enabling effective noise perception and significant quality enhancement of low-field images. In addition, Fréchet inception distance (FID) and learned perceptual image patch similarity (LPIPS) were adopted to evaluate the quality of enhanced low-field MR images, which overcame the challenge of quantifying enhancement performance in low-field MRI images, and enabled a scientific assessment of the efficacy of low-field MRI image enhancement. Results On real low-field data, the dual-path denoising and enhancement network exhibited the best performance in both denoising and detail preservation. When trained on the IXI dataset, the FID and LPIPS values reached 0.472 9 and 149.649 3, respectively when trained on the fastMRI dataset, the corresponding values were 0.473 4 and 148.432 3, respectively. Conclusion The low-field MRI image enhancement method based on adaptive noise perception can effectively enhance low-field images, which promotes the clinical application of low-field MRI devices.
Last Update: 2026-03-27