SwinCT: feature enhancement based low-dose CT images denoising with swin transformer

• Published in: Multimedia systems Vol. 30; no. 1; p. 1
• Main Authors: Jian, Muwei, Yu, Xiaoyang, Zhang, Haoran, Yang, Chengdong
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2024; Springer Nature B.V
• Subjects: Algorithms; Computed tomography; Computer Communication Networks; Computer Graphics; Computer Science; Cryptology; Data Storage Representation; Deep learning; Encoders-Decoders; Image degradation; Image quality; Medical imaging; Medical research; Multimedia Information Systems; Neural networks; Noise; Noise reduction; Operating Systems; Performance measurement; Radiation; Regular Paper; X-rays; Images denoising; Transformer; Pre-training model; Low-dose CT
• ISSN: 0942-4962; 1432-1882
• DOI: 10.1007/s00530-023-01202-x

A bstract To reduce the potential harm to patients from X-ray radiation in computed tomography (CT), low-dose ray CT (LDCT) was conspicuous in clinical diagnosis and evaluation. However, the excessive noises in the LDCT scan significantly degrades the image quality, which seriously affects the clinical diagnostic efficacy. In this paper, we propose SwinCT, a feature-enhanced model for LDCT images noise reduction. SwinCT employs the feature enhancement module (FEM) based on Swin Transformer to extract and augment the high-level features of medical images, and simultaneously combines with the deep noise reduction encoder-decoder network in the downstream task, thus ensuring that more tissue and lesion details are retained after images denoising. Compared with the original LDCT images of noisy surrounding, the denoised image quality is significantly improved by the devised SwinCT denoising model, and the performance metrics of our method are also competitive with other advanced LDCT image denoising methods.