대한방사선기술학회지:방사선기술과학 (Journal of radiological science and technology)

  • 제47권1호
  • /
  • Pages.29-37
  • /
  • 2024
  • /
  • 2288-3509(pISSN)
  • /
  • 2384-1168(eISSN)
대한방사선과학회 (Korean Society of Radiological Science)
권호 표지
DOI QR코드

DOI QR Code

갑상선 초음파 영상의 평활화 알고리즘에 따른 U-Net 기반 학습 모델 평가

Evaluation of U-Net Based Learning Models according to Equalization Algorithm in Thyroid Ultrasound Imaging

  • 정무진 (세브란스병원 핵의학과) ;
  • 오주영 (연세암병원 방사선종양학과) ;
  • 박훈희 (신구대학교 방사선학과) ;
  • 이주영 (신구대학교 방사선학과)
  • Moo-Jin Jeong (Department of Nuclear Medicine, Severance Hospital) ;
  • Joo-Young Oh (Department of Radiation Oncology, Yonsei Cancer Center) ;
  • Hoon-Hee Park (Department of Radiological Technology, Shingu College) ;
  • Joo-Young Lee (Department of Radiological Technology, Shingu College)
  • 투고 : 2023.10.09
  • 심사 : 2024.01.24
  • 발행 : 2024.02.28
PDF 다운로드
⟨ 이전 논문 다음 논문 ⟩

초록

This study aims to evaluate the performance of the U-Net based learning model that may vary depending on the histogram equalization algorithm. The subject of the experiment were 17 radiology students of this college, and 1,727 data sets in which the region of interest was set in the thyroid after acquiring ultrasound image data were used. The training set consisted of 1,383 images, the validation set consisted of 172 and the test data set consisted of 172. The equalization algorithm was divided into Histogram Equalization(HE) and Contrast Limited Adaptive Histogram Equalization(CLAHE), and according to the clip limit, it was divided into CLAHE8-1, CLAHE8-2. CLAHE8-3. Deep Learning was learned through size control, histogram equalization, Z-score normalization, and data augmentation. As a result of the experiment, the Attention U-Net showed the highest performance from CLAHE8-2 to 0.8355, and the U-Net and BSU-Net showed the highest performance from CLAHE8-3 to 0.8303 and 0.8277. In the case of mIoU, the Attention U-Net was 0.7175 in CLAHE8-2, the U-Net was 0.7098 and the BSU-Net was 0.7060 in CLAHE8-3. This study attempted to confirm the effects of U-Net, Attention U-Net, and BSU-Net models when histogram equalization is performed on ultrasound images. The increase in Clip Limit can be expected to increase the ROI match with the prediction mask by clarifying the boundaries, which affects the improvement of the contrast of the thyroid area in deep learning model learning, and consequently affects the performance improvement.

키워드

과제정보

This work was supported by the research fund of Shingu College

참고문헌

  1. Todsen T. Ultrasound-guided fine-needle aspiration biopsy of thyroid nodules. Head & Neck. 2021;43(3):1009-13. DOI: https://doi.org/10.1002/hed.26598 
  2. Ma X, Zhang L. Diagnosis of thyroid nodules based on image enhancement and deep neural networks. Computational Intelligence and Neuroscience. 2022;2022:5582029. DOI: https://doi.org/10.1155/2022/5582029 
  3. Choi YH. U-net based transfer learning model for lesion area segmentation on the breast ultrasound images by optimization of encoder. Kyungpook National University Graduate School; 2022. 
  4. Shin SY, Lee SH, Han HH. A study on residual u-net for semantic segmentation based on deep learning. Journal of Digital Convergence. 2021;19(6):251-8. DOI: https://doi.org/10.14400/JDC.2021.19.6.251 
  5. Lee J. Using deep learning for image-based superficial femoral artery disease detection. Catholic University Graduate School; 2022. 
  6. Kim YJ, Park YR, Kim YJ, Ju W, Nam K, Kim KG. A performance comparison of histogram equalization algorithms for cervical cancer classification model. Journal of Biomedical Engineering Research. 2021;42(3):80-5. DOI: https://kiss.kstudy.com/DetailOa/Ar?key=52271802# 
  7. Yoshimi Y, Mine Y, Ito S, et al. Image preprocessing with contrast-limited adaptive histogram equalization improves the segmentation performance of deep learning for the articular disk of the temporomandibular joint on magnetic resonance images. Oral Surg Oral Med Oral Pathol Oral Radiol. 2023. DOI: https://doi.org/10.1016/j.oooo.2023.01.016 
  8. Cho YH. Quality enhancement of medical images by using nonlinear histogram equalization function. Journal of the Korean Society of Industry Convergence. 2010;13(1):23-30. DOI: https://doi.org/10.5391/JKIIS.2010.20.1.030 
  9. Zuiderveld K. Contrast limited adaptive histogram equalization. In: Heckbert P. ed. Graphics gems IV. Academic Press; 1994. DOI: https://doi.org/10.1016/B978-0-12-336156-1.50061-6 
  10. https://www.deepphi.ai/ 
  11. Son JW, Moon GS, Kim Y. Automatic detection system of underground pipe using 3D GPR exploration data and deep convolutional neural networks. Journal of the Korea Society of Computer and Information. 2021;26(2):27-37. DOI: https://doi.org/10.9708/jksci.2021.26.02.027 
  12. Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation. DOI: https://doi.org/10.48550/arXiv.1411.4038 
  13. Lee JH, Lee DH, Wang WJ, et al. Left atrial segmentation technique based on u-net. Korea Institute of Communication Sciences; 2022:139-40. 
  14. Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation. MICCAI. 2015:9351. DOI: https://doi.org/10.48550/arXiv.1505.04597 
  15. Shin SY, Lee SH, Han HH. Atrous residual u-net for semantic segmentation in street scenes based on deep learning. Journal of Convergence for Information Technology. 2021:28;11(10):45-52. DOI: https://doi.org/10.22156/CS4SMB.2021.11.10.045 
  16. Hwang DH, Moon GS, Kim Y. SKU-net: Improved U-net using selective kernel convolution for retinal vessel segmentation. Journal of the Korea Society of Computer and Information. 2021;26(4):29-37. DOI: https://doi.org/10.9708/JKSCI.2021.26.04.029 
  17. Kim IK, Kim BM, Woo SH, Gwak JH. Contactless user identification system using multi-channel palm images facilitated by triple attention u-net and CNN classifier ensemble models. Journal of the Korea Society of Computer and Information. 2022;27(3):33-43. DOI: https://doi.org/10.9708/jksci.2022.27.03.033 
  18. Shin HS, Song SH, Lee DH, Park JH. Application and evaluation of the attention u-net using UAV imagery for corn cultivation field extraction. Ecology and Resilient Infrastructure. 2021 Dec 31;8(4):253-65. DOI: https://doi.org/10.17820/eri.2021.8.4.253 
  19. Oh JY, Jeong EH, Lee JY, et al. Evaluating usefulness of deep learning based left ventricle segmentation in cardiac gated blood pool scan. Journal of Radiological Science and Technology. 2022;45(2):151-8. DOI: https://doi.org/10.17946/JRST.2022.45.2.151 
  20. Hong JY, Park SH, Jeong YJ. Artificial intelligence based medical imaging: An overview. Journal of Radiological Science and Technology. 2020;43(3):195-208. DOI: https://doi.org/10.17946/JRST.2020.43.3.195