May 2005
Volume 46, Issue 13
ARVO Annual Meeting Abstract  |   May 2005
Nonrigid Intra– and Intermodal Registration of Retinal Images
Author Affiliations & Notes
  • C.A. Amstutz
    Department of Ophthalmology, Inselspital, University of Bern, Bern, CH–3010, Switzerland
  • J. Kowal
    Division of Surgical Technology, Maurice E. Mueller Institute, University of Bern, Bern, CH–3010, Switzerland
  • J.G. Garweg
    Department of Ophthalmology, Inselspital, University of Bern, Bern, CH–3010, Switzerland
  • Footnotes
    Commercial Relationships  C.A. Amstutz, None; J. Kowal, None; J.G. Garweg, None.
  • Footnotes
    Support  None.
Investigative Ophthalmology & Visual Science May 2005, Vol.46, 2580. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      C.A. Amstutz, J. Kowal, J.G. Garweg; Nonrigid Intra– and Intermodal Registration of Retinal Images . Invest. Ophthalmol. Vis. Sci. 2005;46(13):2580.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Abstract: : Purpose:Registration refers to the spatial alignment of patient data from different image modalities (intermodal registration), or from data of the same modality recorded at different times or where spatial overlapping is incomplete (intramodal registration). For an ongoing research project that aims at the quantitative evaluation of the treatment effect of photodynamic therapy, a novel algorithm for the registration of composite fundus photographs and fluorescein angiography images has been developed. Methods: A registration algorithm was used that incorporates local rather than global distortion, the distortion transformation being a bivariate uniform cubic B–spline function. In order to fit the function to the constraints given by the correspondence of extracted features, a multi–level B–spline function hierarchy is used. To use the algorithm for both feature and pixel based similarity measures, the control lattice points are calculated either from feature constraints or by the minimization of a pixel based similarity measure. For intramodal registration, the similarity measure used is the sum of squared intensity differences, for intermodal registration, mutual information is applied. Results:The algorithm is suitable for the fine–adjustment of the registration, achieving sub–pixel accuracy. For a rough initial alignment, feature extraction (optic disc, vessels and their branching points) is used to compute an affine transformation by an iterative closest point algorithm. For the application of pixel based similarity measures such as mutual information, the transformed image has been resampled in a new pixel domain. Various kernels for resampling functions have been applied and compared with respect to intra– and intermodal pixel based registration. Conclusions:We show that the algorithm is efficient and accurate for feature based registration. We show that for pixel based registration, the efficiency and sub–pixel accuracy is highly dependent on the resampling method applied to intermediate transformation stages of the image. Using a local distortion function for registration rather than a global one has several advantages. No assumptions about the global distortion of the images are necessary. Local differences in the amount of distortion are corrected without affecting overall accuracy of the registration. Computational efficiency is high, as the number of parameters that have to be estimated for a local image patch is limited.

Keywords: imaging/image analysis: clinical • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • retina 

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.