Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network
Stationary inverse-geometry digital tomosynthesis (s-IGDT) causes truncation artifacts in reconstructed images due to its geometric characteristics. This study introduces a deep convolutional generative adversarial network (DCGAN)-based out-painting method for mitigating truncation artifacts in s-IG...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-07-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/15/14/7699 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1839616646420889600 |
|---|---|
| author | Burnyoung Kim Seungwan Lee |
| author_facet | Burnyoung Kim Seungwan Lee |
| author_sort | Burnyoung Kim |
| collection | DOAJ |
| description | Stationary inverse-geometry digital tomosynthesis (s-IGDT) causes truncation artifacts in reconstructed images due to its geometric characteristics. This study introduces a deep convolutional generative adversarial network (DCGAN)-based out-painting method for mitigating truncation artifacts in s-IGDT images. The proposed network employed an encoder–decoder architecture for the generator, and a dilated convolution block was added between the encoder and decoder. A dual-discriminator was used to distinguish the artificiality of generated images for truncated and non-truncated regions separately. During network training, the generator was able to selectively learn a target task for the truncated regions using binary mask images. The performance of the proposed method was compared to conventional methods in terms of signal-to-noise ratio (SNR), normalized root-mean-square error (NRMSE), peak SNR (PSNR), and structural similarity (SSIM). The results showed that the proposed method led to a substantial reduction in truncation artifacts. On average, the proposed method achieved 62.31, 16.66, and 14.94% improvements in the SNR, PSNR, and SSIM, respectively, compared to the conventional methods. Meanwhile, the NRMSE values were reduced by an average of 37.22%. In conclusion, the proposed out-painting method can offer a promising solution for mitigating truncation artifacts in s-IGDT images and improving the clinical availability of the s-IGDT. |
| format | Article |
| id | doaj-art-e34d80b3ff7c46069989f342414dd7b5 |
| institution | Matheson Library |
| issn | 2076-3417 |
| language | English |
| publishDate | 2025-07-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Applied Sciences |
| spelling | doaj-art-e34d80b3ff7c46069989f342414dd7b52025-07-25T13:12:05ZengMDPI AGApplied Sciences2076-34172025-07-011514769910.3390/app15147699Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial NetworkBurnyoung Kim0Seungwan Lee1Deepnoid Inc., 33 Digital-ro, Seoul 08377, Republic of KoreaDepartment of Radiological Science, Konyang University, 158 Gwanjeodong-ro, Daejeon 35365, Republic of KoreaStationary inverse-geometry digital tomosynthesis (s-IGDT) causes truncation artifacts in reconstructed images due to its geometric characteristics. This study introduces a deep convolutional generative adversarial network (DCGAN)-based out-painting method for mitigating truncation artifacts in s-IGDT images. The proposed network employed an encoder–decoder architecture for the generator, and a dilated convolution block was added between the encoder and decoder. A dual-discriminator was used to distinguish the artificiality of generated images for truncated and non-truncated regions separately. During network training, the generator was able to selectively learn a target task for the truncated regions using binary mask images. The performance of the proposed method was compared to conventional methods in terms of signal-to-noise ratio (SNR), normalized root-mean-square error (NRMSE), peak SNR (PSNR), and structural similarity (SSIM). The results showed that the proposed method led to a substantial reduction in truncation artifacts. On average, the proposed method achieved 62.31, 16.66, and 14.94% improvements in the SNR, PSNR, and SSIM, respectively, compared to the conventional methods. Meanwhile, the NRMSE values were reduced by an average of 37.22%. In conclusion, the proposed out-painting method can offer a promising solution for mitigating truncation artifacts in s-IGDT images and improving the clinical availability of the s-IGDT.https://www.mdpi.com/2076-3417/15/14/7699stationary inverse-geometry digital tomosynthesistruncation artifactdeep convolutional generative adversarial networksout-painting |
| spellingShingle | Burnyoung Kim Seungwan Lee Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network Applied Sciences stationary inverse-geometry digital tomosynthesis truncation artifact deep convolutional generative adversarial networks out-painting |
| title | Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network |
| title_full | Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network |
| title_fullStr | Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network |
| title_full_unstemmed | Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network |
| title_short | Truncation Artifact Reduction in Stationary Inverse-Geometry Digital Tomosynthesis Using Deep Convolutional Generative Adversarial Network |
| title_sort | truncation artifact reduction in stationary inverse geometry digital tomosynthesis using deep convolutional generative adversarial network |
| topic | stationary inverse-geometry digital tomosynthesis truncation artifact deep convolutional generative adversarial networks out-painting |
| url | https://www.mdpi.com/2076-3417/15/14/7699 |
| work_keys_str_mv | AT burnyoungkim truncationartifactreductioninstationaryinversegeometrydigitaltomosynthesisusingdeepconvolutionalgenerativeadversarialnetwork AT seungwanlee truncationartifactreductioninstationaryinversegeometrydigitaltomosynthesisusingdeepconvolutionalgenerativeadversarialnetwork |