Current Result Document : 5 / 5
| 한글제목(Korean Title) |
손실함수에 따른 적대적 생성 신경망 기반 영상 변환 모델의 생체분포 변화 평가 |
| 영문제목(English Title) |
Evaluation of Image-to-image Translation for Biodistribution Change Using Generative Adversarial Network with Loss Function |
| 저자(Author) |
김강산
변병현
김병철
임상무
우상근
Kangsan Kim
Byung Hyun Byun
Byung-Chul Kim
Sang Moo Lim
Sang-Keun Woo
|
| 원문수록처(Citation) |
VOL 49 NO. 01 PP. 0657 ~ 0659 (2022. 06) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
Image-to-image translation based on the conditional generative adversarial network has been widely used in medical imaging for various purposes. Since GAN framework is known to suffer the instability issue during training, it is important to select the objective function to avoid the problems such as gradient vanishing. In this study, we compared the performance of the models trained with various combinations of the objective functions, in order to translate the distribution of the 18F-FDG over time. According to the training results, using Wasserstein distance made the test dataset more similar to its ground-truth, while using adversarial loss of vanilla GAN worsened the image. Among the objective functions, we verified that the weighted sum of the Wasserstein distance and perceptual loss via VGG19 network showed finest performance. |
| 키워드(Keyword) |
|
| 파일첨부 |
PDF 다운로드
|
