U-Net 모델을 이용한 안저 이미지 재생성에 관한 연구
A Study on the Reproduction of Fundus Image Using U-Net Model
0
조회
0
다운로드
2
인용
0
참고문헌
키워드
초록
의료 분야에서는 인공지능의 영상처리기술을 바탕으로 의료 영상 데이터를 활용한 연구가 활발히 이루어지고 있다. 하지만 인공지능을 활용한 의료 분야 연구는 의료데이터를 획득하는 것에 큰 어려움이 따른다. 의료 데이터를 사용하여 연구를 진행하기 위해서는 IRB(Institutional Review Board)를 비롯한 많은 법적 절차를 거쳐야 하며, 의료인의 데이터 검증 절차를 거쳐야 한다. 이는 연구를 진행하는 것에 있어 많은 시간적, 경제적 손실을 발생시킨다. 따라서 이러한 문제점을 해결하고자 본 논문에서는 의료영상 데이터 중 안저 이미지를 재생성하는 연구를 진행하였다. 본 연구에서 사용한 데이터는 안저 이미지이며, 정상, 녹내장, 당뇨성 망막병증, 연령관련 황반변성으로 구성하였다. 딥러닝 네트워크인 U-Net을 사용하여 안저 이미지를 재생성하였으며, 재생성된 데이터는 이미지 유사도 평가 및 의료인을 통해 검증하였다. 재생성 결과 원본 데이터와 시각적으로 유사한 데이터를 생성하였다. 또한 의료인의 검증 결과 재생성된 데이터에 대해 20~30%는 찾아내지 못할 정도로 재생성 모델의 성능은 뛰어났다. 향후 연구로는 전이 학습을 통해 재생성 모델을 개선할 것이다. 또한 재생성 데이터의 악용을 방지하기 위해 재생성 데이터 검출 모델을 연구할 계획이다.
Abstract
In the medical field, research using medical image data is being actively conducted based on artificial intelligence image processing technology. However, research in the medical field using artificial intelligence has great difficulties in obtaining medical data. In order to conduct research using medical data, it must go through many legal procedures, including IRB(Institutional Review Board), and must go through data verification procedures by medical personnel. This causes a lot of time and economic losses in conducting research. Therefore, in order to solve this problem, this paper conducted a study to reproduce the fundus image among medical image data. The data used in this study were fundus images, consisting of normal, glaucoma, diabetic retinopathy, and age-related macular degeneration. The fundus image was reproduced using U-Net, a deep learning network, and the reproduced data was verified through image similarity evaluation and medical personnel. As a result of reproduction, data that is visually similar to the original data were reproduce. In addition, the performance of the reproduction model was so higher that 20 to 30% of the reproduced data could not be found as a result of medical personnel's verification. Future research will improve the reproduction model through transfer learning. In addition, we plan to study a reproductive data detection model to prevent abuse of reproductive data.
참고문헌 (18)
- H. Dhamo, "Semantic image manipulation using scene graphs", Proceedings of the IEEE, 2020, 5213~5222
- D. Shen, "Deep learning in medical image analysis", Annual Review of Biomedical Engineering, 19(1), pp.221~248 (2017)
- 김영재, "의료 영상에 최적화된 딥러닝 모델의 개발", 대한영상의학회지, 81(6), pp.1274~1289 (2020), 대한영상의학회[DOI]
- "Institutional Bioethics Committee, Review Process"[URL]
- L. N. Dyrbye, "Medical education research and IRB review: an analysis and comparison of the IRB review process at six institutions", Academic Medicine, 82(7), pp.654~660 (2007)
- C. Sun, "Revisiting unreasonable effectiveness of data in deep learning era", Proceedings of the IEEE international conference on computer vision, pp.843~852 (2017)
- L. Ying, "A wavelet based image sharpening algorithm", 2008International Conference on Computer Science and Software Engineering, 1, pp.1053~1056 (2008)
- S. Y. Wang, "Cnn-generated images are surprisingly easy to spot... for now", Proceedings of the IEEE, 2020, 8695~8704
- O. Ronneberger, "U-net:Convolutional networks for biomedical image segmentation", International Conference on Medical image computing and computer-assisted intervention, pp.234~241 (2015)
- N. Siddique, "U-net and its variants for medical image segmentation: A review of theory and applications", IEEE Access, 9, pp.82031~82057 (2021)
- I. Goodfellow, "Generative adversarial net", Proc. NeurIPS, pp.2672~2680 (2014)
- Y. J. Jo, "Research trends of generative adversarial networks and image generation and translation", ETRI Electronic Communication Trend Analysis, 35(4), pp.91~102 (2020)
- K. B. Park, "M-GAN:Retinal blood vessel segmentation by balancing losses through stacked deep fully convolutional networks", IEEE Access, 8, pp.146308~146322 (2020)
- A. Mason, "Comparison of objective image quality metrics to expert radiologists’ scoring of diagnostic quality of MR images", IEEE transactions on medical imaging, 39(4), pp.1064~1072 (2019)
- U. Sara, "Image quality assessment through FSIM, SSIM, MSE and PSNR—a comparative study", Journal of Computer and Communications, 7(3), pp.8~18 (2019)
- G. Parmar, "On buggy resizing libraries and surprising subtleties in fid calculation" (2021)
- "Kaggle, Ocular Disease Recognition: Right and left eye fundus photographs of 5000 patients"[URL]
- "resource book of the results of the 2018 National Health and Nutrition Survey and Youth Health Behavior Survey"[URL]