Modularization of Deep Networks Allows Cross-Modality Reuse

Lesson Learnt
  • Weilin FuEmail author
  • Lennart Husvogt
  • Stefan Ploner
  • James G. Fujimoto
  • Andreas Maier
Conference paper
Part of the Informatik aktuell book series (INFORMAT)


Fundus photography and Optical Coherence Tomography Angiography (OCT-A) are two commonly used modalities in ophthalmic imaging. With the development of deep learning algorithms, fundus image processing, especially retinal vessel segmentation, has been extensively studied. Built upon the known operator theory, interpretable deep network pipelines with well-defined modules have been constructed on fundus images. In this work, we firstly train a modularized network pipeline for the task of retinal vessel segmentation on the fundus database DRIVE. The pretrained preprocessing module from the pipeline is then directly transferred onto OCT-A data for image quality enhancement without further fine-tuning. Output images show that the preprocessing net can balance the contrast, suppress noise and thereby produce vessel trees with improved connectivity in both image modalities. The visual impression is confirmed by an observer study with five OCT-A experts. Statistics of the grades by the experts indicate that the transferred module improves both the image quality and the diagnostic quality. Our work provides an example that modules within network pipelines that are built upon the known operator theory facilitate cross-modality reuse without additional training or transfer learning.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    Srinidhi CL, Aparna P, Rajan J. Recent advancements in retinal vessel segmentation. J med Syst. 2017;41(4):70.Google Scholar
  2. 2.
    Husvogt L, Ploner S, Maier A. Optical coherence tomography. Springer, Cham; 2018. p. 251–261.Google Scholar
  3. 3.
    Choi W, Moult EM, Waheed NK, et al. Ultrahigh-Speed, swept-source optical coherence tomography angiography in nonexudative age-related macular degeneration with geographic atrophy. Ophthalmology. 2015;.Google Scholar
  4. 4.
    Maier AK, Syben C, Stimpel B, et al. Learning with known operators reduces maximum error bounds. Nature machine intelligence. 2019;1(8):373–380.Google Scholar
  5. 5.
    Fu W, Breininger K, Schaffert R, et al. A divide-and-conquer approach towards understanding deep networks. In: MICCAI. Springer; 2019. p. 183–191.Google Scholar
  6. 6.
    Lin TY, Goyal P, Girshick R, et al. Focal loss for dense object detection. In: Proceedings of the IEEE international conference on computer vision; 2017. p. 2980–2988.Google Scholar
  7. 7.
    Kingma DP, Ba J. Adam: a method for stochastic optimization. arXiv preprint arXiv:14126980. 2014;.
  8. 8.
    Zuiderveld K. Contrast limited adaptive histogram equalization. In: Graphics gems IV. Academic Press Professional, Inc.; 1994. p. 474–485.Google Scholar
  9. 9.
    Husvogt L, Ploner S, Moult EM, et al. Using medical image reconstruction methods for denoising of OCTA data. Invest Ophthal Vis Sci. 2019;60:3096.Google Scholar

Copyright information

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2020

Authors and Affiliations

  • Weilin Fu
    • 1
    • 2
    Email author
  • Lennart Husvogt
    • 1
    • 4
  • Stefan Ploner
    • 1
  • James G. Fujimoto
    • 4
  • Andreas Maier
    • 1
    • 3
  1. 1.Pattern Recognition LabFriedrich-Alexander UniversityErlangenDeutschland
  2. 2.International Max Planck Research School Physics of Light (IMPRS-PL)ErlangenDeutschland
  3. 3.Erlangen Graduate School in Advanced Optical Technologies(SAOT)ErlangenDeutschland
  4. 4.Biomedical Optical Imaging and Biophotonics GroupMITCambridgeUSA

Personalised recommendations