Neighboring Slice Noise2Noise: Self-Supervised Medical Image Denoising from Single Noisy Image Volume

• DOI: 10.48550/arxiv.2411.10831
• Publication Date: 2024-11-16
• Authors: Langrui Zhou; Ziteng Zhou; Xinyu Huang; Xiangyu Zhang; H. J. Wang; Guang Li
• Keywords: FOS: Computer and information sciences; Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV); Computer Science - Computer Vision and Pattern Recognition; FOS: Electrical engineering, electronic engineering, information engineering; Electrical Engineering and Systems Science - Image and Video Processing

In the last few years, with rapid development of deep learning technologies, supervised methods based on convolutional neural networks have greatly enhanced performance medical image denoising. However, these require large quantities noisy-clean pairs for training, which limits their practicality. Although some researchers attempted to train denoising using only single noisy images, existing self-supervised methods, including blind-spot-based and data-splitting-based heavily rely assumption that noise is pixel-wise independent. this often does not hold in real-world images. Therefore, field imaging, there remains a lack simple practical can achieve high-quality paper, we propose novel method, Neighboring Slice Noise2Noise (NS-N2N). The proposed method utilizes neighboring slices within volume construct weighted training data, then trains network scheme regional consistency loss inter-slice continuity loss. NS-N2N requires obtained from one imaging procedure itself. Extensive experiments demonstrate outperforms state-of-the-art both processing efficiency. Furthermore, since operates solely domain, it free device-specific issues such as reconstruction geometry, making easier apply various clinical practices.