DOI QR코드

DOI QR Code

Effect of Data Augmentation Techniques for Classification of Focal Liver Lesions Based on Deep Convolutional Neural Networks in Abdominal CT Images

복부 CT 영상에서 심층 합성곱 신경망 기반의 국소 간 병변 분류를 위한 데이터 증강 기법의 효과 분석

  • Deokseon Kim (Department of Software Convergence, Seoul Women's University) ;
  • Ahra Woo (Department of Software Convergence, Seoul Women's University) ;
  • Hansang Lee (School of Electrical Engineering, Korea Advanced Institute of Science and Technology) ;
  • Helen Hong (Department of Software Convergence, Seoul Women's University)
  • 김덕선 (서울여자대학교 소프트웨어융합학과) ;
  • 우아라 (서울여자대학교 소프트웨어융합학과) ;
  • 이한상 (한국과학기술원 정보전자연구소) ;
  • 홍헬렌 (서울여자대학교 소프트웨어융합학과)
  • Received : 2022.12.08
  • Accepted : 2023.03.09
  • Published : 2023.06.01

Abstract

In this paper, we analyze the effects in medical images by applying various data augmentation techniquesto deep convolutional neural network learning for classifying focal liver lesions in abdominal CT images. We apply affine transformation-based, StyleGAN, Mixup, and Augmix-based data augmentation techniquesto the VGG16 convolutional neural network, respectively, to learn to classify local liver lesionsinto cysts, hemangiomas, and metastases. For the experiments, we validate and analyze the effect of the data augmentation through both a quantitative assessment by comparing accuracy, sensitivity, and specificity for classification results of models trained by each data augmentation technique and a qualitative assessment by observing augmented image examples and the tSNE feature distributions.

본 논문에서는 복부CT 영상에서 국소 간 병변분류를 위한 심층 합성곱 신경망 학습에 다양한 데이터 증강 기법을 적용하여 의료 영상에서의 효과를 분석한다. 이를 위해 어파인 변환 기반, StyleGAN, Mixup, Augmix 기반 데이터 증강 기법을 각각 VGG16 심층 합성곱 신경망에 적용하여 국소간병변을 낭종, 혈관종, 전이암으로 분류하는 학습을 수행한다. 실험을 위해 각 데이터 증강 기법에 의해 훈련된 모델의 분류 결과에 대하여 정확도, 민감도, 특이도 분석을 통해 정량적 평가를 진행하고, 증강 영상 예시 분석과 tSNE 분석을 통해 정성적 평가를 수행하여 데이터 증강 기법의 효과를 분석한다.

Keywords

Acknowledgement

논문에서 사용한 복부 CT 영상 데이터를 제공해주신 세브란스병원 영상의학과 임준석 교수님께 감사의 말씀을 전합니다. 이 논문은 서울여자대학교 학술연구비의 지원에 의한 것임(2023-0092).

References

  1. P. R. Ros, and K. J. Mortele, "Diffuse liver disease." Clinics in liver disease, pp. 181-201, 2002.
  2. S. Manfredi, C. Lepage, C. Hatem, O. Coatmeur, J. Faivre, and AM. Bouvier, "Epidemiology and management of liver metastases from colorectal cancer," Ann Surg 244, pp. 254-259, 2006. https://doi.org/10.1097/01.sla.0000217629.94941.cf
  3. G. Bengtsson, G. Carlsson, L. Hafstrom, and P. E. Jonsson, "Natural. history of patients with untreated liver metastases from colorectal cancer." The American Journal of Surgery, pp. 586-589, 1981.
  4. M. Nino-Murcia, E.W. Olcott, R.B. Jeffrey, R.L. Lamm, C.F. Beaulieu, and K.A. Jain, "Focal liver lesions: pattern-based classification scheme for enhancement at arterial phase CT," Radiology, 215, pp. 746-751, 2000. https://doi.org/10.1148/radiology.215.3.r00jn03746
  5. C.C. Chang, H.H. Chen, Y.C. Chang, M.Y. Yang, C.M. Lo, W.C. Ko, Y.F. Lee, K.L. Liu, and R.F. Chang, "Computer-aided diagnosis of liver tumors on computed tomography images," Comput Methods Prog Biomed, 145, pp. 45-51, 2017. https://doi.org/10.1016/j.cmpb.2017.04.008
  6. A. Radford, L. Metz, and S. Chintala, "Unsupervised representation learning with deep convolutional generative adversarial networks." International Conference on Learning Representations(ICLR), 2016.
  7. Karras, T., Aila, T., Laine, S., Lehtinen, J., "Progressive growing of GANs for improved quality, stability, and variation." ICLR, 2017.
  8. Karras, T., Laine, S., and Aila, T., "A style-based generator. architecture for generative adversarial networks." CVPR, 2019.
  9. H. Salehinejad, S. Valaee, T. Dowdell, E. Colak, and J. Barfett, "Generalization of deep neural networks for chest pathology classification in x-rays using generative adversarial networks." IEEE international conference on acoustics, speech and signal processing (ICASSP), 2018.
  10. M. Frid-Adar, I. Diamant, E. Klang, M. Amitai, J. Goldberger, and H. Greenspan "Gan-based synthetic medical image augmentation for increased cnn performance in liver lesion classification." Neurocomputing, vol. 321, pp. 321-331, 2018. https://doi.org/10.1016/j.neucom.2018.09.013
  11. D. Zhao, D. Zhu, J. Lu, Y. Luo, and G. Zhang, "Synthetic medical images using F&BGAN for improved lung nodules classification by multi-scale VGG16." Symmetry, vol. 10, no. 10, p. 519, 2018.
  12. P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, "Imageto-image translation with conditional adversarial networks." CVPR, 2017.
  13. Lee, H., Lee, H., Hong, H., Bae, H., Lim, J. S., and Kim, J, "Classification of focal liver lesions in CT images using convolutional neural networks with lesion information augmented patches and synthetic data augmentation." Medical physics 48.9, pp. 5029-5046, 2021. https://doi.org/10.1002/mp.15118
  14. Zhang, H., Cisse, M., Dauphin, Y. N. and Lopez-Paz, D. "mixup: Beyond empirical risk minimization." International Conference on Learning Representations(ICLR), 2017.
  15. Hendrycks, D., Mu, N., Cubuk, E. D., Zoph, B., Gilmer, J., and Lakshminarayanan, B. "AugMix: A simple data processing method to improve robustness and uncertainty" International Conference on Learning Representations(ICLR), 2019.
  16. Yun, S., Han, D., Oh, S. J., Chun, S., Choe, J., and Yoo, Y, "Cutmix: Regularization strategy to train strong classifiers with localizable features." Proceedings of the IEEE/CVF international conference on computer vision, 2019.
  17. M. Nishio, S. Noguchi, H. Matsuo, and T. Murakami, "Automatic classification between COVID-19 pneumonia, non-COVID-19 pneumonia, and the healthy on chest X-ray image: combination of data augmentation methods." Scientific reports 10.1, pp. 1-6., 2020. https://doi.org/10.1038/s41598-019-56847-4
  18. Rajan, D., Thiagarajan, J. J., Karargyris, A. and Kashyap, S. "Self-training with improved regularization for sample-efficient chest x-ray classification." Medical Imaging 2021: Computer-Aided Diagnosis. SPIE, Volume 11597, 2021.
  19. Ozdemir, O. and Sonmez, E. B. "Attention mechanism and mixup data augmentation for classification of COVID-19 Computed Tomography images." Journal of King Saud University -Computer and Information Sciences, Volume 34, Issue 8, Part B, Pages 5587-6504, 2022. https://doi.org/10.1016/j.jksuci.2021.02.009
  20. van der Maaten L, and Hinton G, "Visualizing data using t-sne."Journal of machine learning research, vol 9, pp. 2579-2605, 2008.
  21. R. Hao, K. Namdar, L. Liu, M. A. Haider, and F. Khalvati, "A comprehensive study of data augmentation strategies for prostate cancer detection in diffusion-weighted MRI using convolutional neural networks." Journal of Digital Imaging, pp. 862-876, 2021.
  22. M. Kim and H.-J. Bae, "Data augmentation techniques for deep learning based medical image analyses." Journal of the Korean Society of Radiology, vol. 81, no. 6, 2020.
  23. H. Bae, H. Lee, S. Kim, K. Han, H. Rhee, D. Kim, H. Kwon, H. Hong, and J.S. Lim, "Radiomics Analysis of Contrast-Enhanced CT for Classification of Hepatic Focal Lesions in Colorectal Cancer Patients: Its Limitations Compared to Radiologists," European Radiology, 31, pp.8786-8796, 2021. https://doi.org/10.1007/s00330-021-07877-y