DC Field | Value | Language |
---|---|---|
dc.contributor.advisor | 조승룡 | - |
dc.contributor.author | Soh, Jeongtae | - |
dc.contributor.author | 소정태 | - |
dc.date.accessioned | 2024-07-26T19:31:36Z | - |
dc.date.available | 2024-07-26T19:31:36Z | - |
dc.date.issued | 2023 | - |
dc.identifier.uri | http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=1052045&flag=dissertation | en_US |
dc.identifier.uri | http://hdl.handle.net/10203/321149 | - |
dc.description | 학위논문(박사) - 한국과학기술원 : 원자력및양자공학과, 2023.8,[v, 58 p. :] | - |
dc.description.abstract | In this study, we developed image reconstruction algorithms for carbon nanotube(CNT) array-based digital tomosynthesis (DTS). Moreover, since this system differs from the conventional DTS, image reconstruction algorithms had to be modified appropriately. Filtered back-projection (FBP) and maximum likelihood exposure maximization (MLEM) are the most commonly used image reconstruction algorithms for DTS. In FBP, we applied the 2D ramp filter, the convolution of 1D filters in each direction, instead of the conventional 1D filter. Furthermore, in both FBP and MLEM cases, the data was multiplied by a smoothed binary mask to remove image artifacts caused by scattering signals generated on the boundary between the part blocked by the collimator mounted in front of the source and the part that was not.After reconstructing the 3D image, we also developed an image processing method to create a 2D synthetic image from the 3D image. This method allows us to produce 3D and 2D images without additional doses. There are numerous algorithms for 2D image synthesis in DTS. However, the characteristics and artifacts of reconstructed images are different using array-based sources, so they are similar to those of optical microscopes rather than conventional DTS. Therefore, we developed an advanced smooth manifold extraction (ASME) in this study by modifying the smooth manifold extraction (SME) used in optical microscopy to suit our system.We also studied a deep learning algorithm that improves the depth resolution of DTS images. Unlike full-3D CT images, a DTS image is called a quasi-3D image due to the poor resolution of the reconstructed image because projection data is obtained only at a limited angle. In order to solve this problem, we used supervised learning called 3D U-net. Furthermore, by studying the loss function that can be applied depending on target objects, we made the method to be used both in medical and industrial fields. | - |
dc.language | eng | - |
dc.publisher | 한국과학기술원 | - |
dc.subject | 디지털 단층영상 합성▼a영상 재구성 알고리즘▼a2D 합성 영상▼a3D U-net | - |
dc.subject | Digital tomosynthesis▼aImage reconstruction algorithm▼a2D synthetic image▼a3D U-net | - |
dc.title | Development of image reconstruction method for carbon nanotube array-based digital tomosynthesis system | - |
dc.title.alternative | 탄소나노튜브 배열 기반 디지털 단층영상 합성 시스템을 위한 영상 재구성 방법 개발 | - |
dc.type | Thesis(Ph.D) | - |
dc.identifier.CNRN | 325007 | - |
dc.description.department | 한국과학기술원 :원자력및양자공학과, | - |
dc.contributor.alternativeauthor | Cho, Seungryong | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.