Limiting the potential risks associated with radiation exposure is critically important when obtaining a diagnostic image. However, lowering the level of radiation may cause excessive noise and artifacts in computed tomography (CT) scans. In this study, we implemented and tested the performance of patch-based and block-based REDCNN models and revealed that a 3D kernel is efficient in removing 3D noise and artifacts. Additionally, we applied a 3D bilateral filter and a 2D-based Landweber iteration method to remove any remaining noise and to prevent the edges from blurring, which are limitations of a deep learning-based noise reduction system. For the 2D-based Landweber iteration, we examined the requisite step size and the number of iterations. The representative CT noise and artifacts, which were Gaussian noise and view aliasing artifacts, respectively, were simulated on XCAT and reproduced in vivo to verify that the proposed method could be used in an analogous clinical setting. Lastly, the performance of the proposed algorithm was evaluated on in vivo data with real low-dose noise. Our proposed method effectively suppressed complex noise without losing diagnostic features in both the simulation study and experimental evaluation. Furthermore, for the simulation study, we adopted a numerical observer model to evaluate the structural fidelity of the image quality more appropriately than existing image quality assessment methods.
|Journal||Machine Vision and Applications|
|Publication status||Published - 2021 Nov|
Bibliographical noteFunding Information:
This work was supported by the National Research Foundation of Korea (NRF) Grant funded by the Korean government (MSIP: Ministry of Science, ICT, and Future Planning) (No. NRF-2020R1A4A1016619, NRF-2020R1F1A1073774), and by the Korea Medical Device Development Fund grant funded by the Korea government (the Ministry of Science and ICT, the Ministry of Trade, Industry and Energy, the Ministry of Health & Welfare, the Ministry of Food and Drug Safety) (Project Number: KMDF_PR_20200901_0016, 9991006689).
© 2021, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.
All Science Journal Classification (ASJC) codes
- Hardware and Architecture
- Computer Vision and Pattern Recognition
- Computer Science Applications