Towards Fully Automatic X-Ray to CT Registration

  • Javier EstebanEmail author
  • Matthias Grimm
  • Mathias Unberath
  • Guillaume Zahnd
  • Nassir Navab
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 11769)


The main challenge preventing a fully-automatic X-ray to CT registration is an initialization scheme that brings the X-ray pose within the capture range of existing intensity-based registration methods. By providing such an automatic initialization, the present study introduces the first end-to-end fully-automatic registration framework. A network is first trained once on artificial X-rays to extract 2D landmarks resulting from the projection of CT-labels. A patient-specific refinement scheme is then carried out: candidate points detected from a new set of artificial X-rays are back-projected onto the patient CT and merged into a refined meaningful set of landmarks used for network re-training. This network-landmarks combination is finally exploited for intraoperative pose-initialization with a runtime of 102 ms. Evaluated on 6 pelvis anatomies (486 images in total), the mean Target Registration Error was \(15.0\pm 7.3\) mm. When used to initialize the BOBYQA optimizer with normalized cross-correlation, the average (± STD) projection distance was \(3.4\pm 2.3\) mm, and the registration success rate (projection distance \(<2.5\%\) of the detector width) greater than \(97\%\).


X-ray to CT Registration Projective geometry Neural network Patient-specific training 


  1. 1.
    Bier, B., et al.: X-ray-transform invariant anatomical landmark detection for pelvic trauma surgery. In: Frangi, A.F., Schnabel, J.A., Davatzikos, C., Alberola-López, C., Fichtinger, G. (eds.) MICCAI 2018. LNCS, vol. 11073, pp. 55–63. Springer, Cham (2018). Scholar
  2. 2.
    Van der Bom, M.J., et al.: Robust initialization of 2D–3D image registration using the projection-slice theorem and phase correlation. Med. Phys. 37(4), 1884–1892 (2010)CrossRefGoogle Scholar
  3. 3.
    Hou, B., et al.: Predicting slice-to-volume transformation in presence of arbitrary subject motion. In: Descoteaux, M., Maier-Hein, L., Franz, A., Jannin, P., Collins, D.L., Duchesne, S. (eds.) MICCAI 2017. LNCS, vol. 10434, pp. 296–304. Springer, Cham (2017). Scholar
  4. 4.
    Li, S., Xu, C., Xie, M.: A robust O(n) solution to the perspective-n-point problem. IEEE Trans. Pat. Anal. Mach. Intel. 34(7), 1444–1450 (2012)CrossRefGoogle Scholar
  5. 5.
    Markelj, P., Tomaževič, D., Likar, B., Pernuš, F.: A review of 3D/2D registration methods for image-guided interventions. Med. Im. Anal. 16(3), 642–661 (2012)CrossRefGoogle Scholar
  6. 6.
    Rackerseder, J., Baust, M., Göbl, R., Navab, N., Hennersperger, C.: Initialize globally before acting locally: enabling landmark-free 3D US to MRI registration. In: Frangi, A.F., Schnabel, J.A., Davatzikos, C., Alberola-López, C., Fichtinger, G. (eds.) MICCAI 2018. LNCS, vol. 11070, pp. 827–835. Springer, Cham (2018). Scholar
  7. 7.
    Roth, H.R., et al.: A new 2.5 D representation for lymph node detection in CT. The Cancer Imaging Archive (2015)Google Scholar
  8. 8.
    Suzuki, S., Abe, K.: Topological structural analysis of digitized binary images by border following. Comput. Vis. Graph. Image Process. 30(1), 32–46 (1985)CrossRefGoogle Scholar
  9. 9.
    Wei, S.E., Ramakrishna, V., Kanade, T., Sheikh, Y.: Convolutional pose machines. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 4724–4732 (2016)Google Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Javier Esteban
    • 1
    Email author
  • Matthias Grimm
    • 1
  • Mathias Unberath
    • 2
  • Guillaume Zahnd
    • 1
  • Nassir Navab
    • 1
    • 3
  1. 1.Computer Aided Medical ProceduresTechnische Universität MünchenGarching bei MünchenGermany
  2. 2.Laboratory for Computational Sensing + RoboticsJohns Hopkins UniversityBaltimoreUSA
  3. 3.Computer Aided Medical ProceduresJohns Hopkins UniversityBaltimoreUSA

Personalised recommendations