Research of MB SENSE DTI Denoising Based on Self-Supervised Learning
CAO Da1, WANG Wei2, XU Lulu1, WANG Chuanbing1, CHEN Wei1, WU Xiaoling2
1. Department of Radiology, The First Affiliated Hospital of Nanjing Medical University, Nanjing Jiangsu 210019, China;
2. School of Biomedical Engineering and Informatics, Nanjing Medical University, Nanjing Jiangsu 211166, China
Abstract：Objective To explore the feasibility of noise denoising of self-supervised learning network Patch2Self in multiband
SENSE diffusion tensor imaging (MB SENSE DTI). Methods Brain MB SENSE (MB factor 4) DTI data from 24 healthy volunteers
by 3T MRI were included in this study. The MB SENSE DTI was denoised using the self-supervised learning network Patch2Self,
non-local means (NLM) and local principal component analysis (LPCA). Objective evaluation was performed using peak signal to
noise ratio (PSNR), structural similarity (SSIM); subjective evaluation of image noise, contrast resolution, overall quality and overall quality of the respective fractional anisotropy (FA) maps by radiologist. Results In terms of PSNR, Patch2Self was better than
NLM but lower than LPCA (P<0.05); in terms of SSIM, Patch2Self was better than NLM and LPCA (P<0.05); in terms of noise,
Patch2Self was significantly better than NLM (P<0.05), while there were not statistically difference between Patch2Self and
LPCA images quality (P>0.05); in terms of contrast resolution, overall quality, and overall quality of FA maps, Patch2Self was
superior than NLM and LPCA (P<0.05). Conclusion The self-supervised learning network Patch2Self can denoise MB SENSE DTI
images, preserve the brain tissue structure and improve the image quality.
曹达1，王伟2，徐露露1，王传兵1，陈巍1，吴小玲2. 基于自监督学习网络的多层同时扫描扩散张量成像去噪研究[J]. 中国医疗设备, 2022, 37(8): 142-146.
CAO Da1, WANG Wei2, XU Lulu1, WANG Chuanbing1, CHEN Wei1, WU Xiaoling2. Research of MB SENSE DTI Denoising Based on Self-Supervised Learning. China Medical Devices, 2022, 37(8): 142-146.
