Advertisement

Domain Adaptation for Deviating Acquisition Protocols in CNN-Based Lesion Classification on Diffusion-Weighted MR Images

  • Jennifer KamphenkelEmail author
  • Paul F. Jäger
  • Sebastian Bickelhaupt
  • Frederik Bernd Laun
  • Wolfgang Lederer
  • Heidi Daniel
  • Tristan Anselm Kuder
  • Stefan Delorme
  • Heinz-Peter Schlemmer
  • Franziska König
  • Klaus H. Maier-Hein
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 11040)

Abstract

End-to-end deep learning improves breast cancer classification on diffusion-weighted MR images (DWI) using a convolutional neural network (CNN) architecture. A limitation of CNN as opposed to previous model-based approaches is the dependence on specific DWI input channels used during training. However, in the context of large-scale application, methods agnostic towards heterogeneous inputs are desirable, due to the high deviation of scanning protocols between clinical sites. We propose model-based domain adaptation to overcome input dependencies and avoid re-training of networks at clinical sites by restoring training inputs from altered input channels given during deployment. We demonstrate the method’s significant increase in classification performance and superiority over implicit domain adaptation provided by training-schemes operating on model-parameters instead of raw DWI images.

Keywords

Convolutional neural networks Diffusion-weighted MR imaging Deep learning Lesion classification Domain adaptation 

References

  1. 1.
    Lauby-Secretan, B.: Breast-cancer screening-viewpoint of the IARC working group. New Engl. J. Med. 372(24), 2353–2358 (2015)CrossRefGoogle Scholar
  2. 2.
    Wu, D.: Characterization of breast tumors using diffusion kurtosis imaging (DKI). PloS One 9(11), e113240 (2014)CrossRefGoogle Scholar
  3. 3.
    Sun, K.: Breast cancer: diffusion kurtosis MR imaging diagnostic accuracy and correlation with clinical-pathologic factors. Radiology 277(1), 4655 (2015)CrossRefGoogle Scholar
  4. 4.
    Jäger, P.F., et al.: Revealing hidden potentials of the q-Space signal in breast cancer. In: Descoteaux, M., Maier-Hein, L., Franz, A., Jannin, P., Collins, D.L., Duchesne, S. (eds.) MICCAI 2017. LNCS, vol. 10433, pp. 664–671. Springer, Cham (2017).  https://doi.org/10.1007/978-3-319-66182-7_76CrossRefGoogle Scholar
  5. 5.
    Jäger, P.F., et al.: Complementary value of end-to-end deep learning and radiomics in breast cancer classification on diffusion-weighted MR. In: ISMRM (2017)Google Scholar
  6. 6.
    Ghodrati, M., et al.: Feedforward object-vision models only tolerate small image variations compared to human. Front. Comput. Neurosci. 8, 74 (2014)Google Scholar
  7. 7.
    Nie, D., et al.: Medical image synthesis with context-aware generative adversarial networks. In: MICCAI, pp. 417–425 (2017)CrossRefGoogle Scholar
  8. 8.
    Isola, P., et al.: Image-to-image translation with conditional adversarial networks. In: IEEE Conference on CVPR, p. 5967 (2017)Google Scholar
  9. 9.
    Rezende, D., Jimenez, S.M., Wierstra, D.: Stochastic backpropagation and approximate inference in deep generative models. ICML 32(2), 1278–1286 (2014)Google Scholar
  10. 10.
    Kingma, D., Welling, M.: Auto-encoding variational bayes. In: ICLR (2014)Google Scholar
  11. 11.
    Havaei, M., et al.: HeMIS: Hetero-modal image segmentation. In: MICCAI, pp. 469–477 (2016)Google Scholar
  12. 12.
    Balleyguier, A.C., et al.: \(\text{ BI-RADS }^{\rm TM}\) classification in mammography. Eur. J. Radiol. 61(2), 192–194 (2007)CrossRefGoogle Scholar
  13. 13.
    Jensen, J.H., et al.: Diffusional kurtosis imaging: the quantification of nongaussian water diffusion by means of magnetic resonance imaging. Magn. Reson. Med. 53(6), 1432–1440 (2005)CrossRefGoogle Scholar
  14. 14.
    Bickelhaupt, S., et al.: Radiomics based on adapted diffusion kurtosis imaging helps to clarify most mammographic findings suspicious for cancer. Radiology 287(3), 761–770 (2018)CrossRefGoogle Scholar
  15. 15.
    Roethke, M.C., et al.: Evaluation of diffusion kurtosis imaging versus standard diffusion imaging for detection and grading of peripheral zone prostate cancer. Invest. Radiol. 50(8), 483–489 (2015)CrossRefGoogle Scholar

Copyright information

© Springer Nature Switzerland AG 2018

Authors and Affiliations

  • Jennifer Kamphenkel
    • 1
    Email author
  • Paul F. Jäger
    • 1
  • Sebastian Bickelhaupt
    • 2
  • Frederik Bernd Laun
    • 2
    • 3
  • Wolfgang Lederer
    • 4
  • Heidi Daniel
    • 5
  • Tristan Anselm Kuder
    • 6
  • Stefan Delorme
    • 2
  • Heinz-Peter Schlemmer
    • 2
  • Franziska König
    • 2
  • Klaus H. Maier-Hein
    • 1
  1. 1.Division of Medical Image ComputingGerman Cancer Research Center (DKFZ)HeidelbergGermany
  2. 2.Department of RadiologyDKFZHeidelbergGermany
  3. 3.Institute of RadiologyUniversity Hospital ErlangenErlangenGermany
  4. 4.Radiological Practice at the ATOS ClinicHeidelbergGermany
  5. 5.Radiology Center Mannheim (RZM)MannheimGermany
  6. 6.Medical Physics in RadiologyDKFZHeidelbergGermany

Personalised recommendations