Denoising of pediatric low dose abdominal CT using deep learning based algorithm

• DOI: 10.1371/journal.pone.0260369
• Publication Date: 2022-01-21T18:56:02Z
• Authors: Hyoung Suk Park; Kiwan Jeon; JeongEun Lee; Sun Kyoung You
• Keywords: 03 medical and health sciences; 0302 clinical medicine; Science; Q; R; Medicine; Research Article; 3. Good health

Objectives To evaluate standard dose-like computed tomography (CT) images generated by a deep learning method, trained using unpaired low-dose CT (LDCT) and standard-dose (SDCT) images. Materials methods LDCT (80 kVp, 100 mAs, n = 83) SDCT (120 200 42) were divided into training (42 42 SDCT) validation (41 LDCT) sets. A generative adversarial network framework was used to train datasets. The method virtual (VIs) from the original (OIs). test proposed 262 33) collected another scanner iterative reconstruction (IR). Image analyses performed qualities of VIs in set compare performance IR set. Results noise lowest both sets (all p<0.001). mean number for portal vein liver lower than that OIs p<0.001) similar those SDCT. contrast-to-noise ratio signal-to-noise (SNR) higher p<0.05). SNR highest among three Conclusion datasets could reduce showed SAFIRE. It can be applied older scanners without IR.