August 2019
Volume 60, Issue 11
Open Access
ARVO Imaging in the Eye Conference Abstract  |   August 2019
Motion correction in 3D-OCT data by intensity-based image registration: an evaluation study
Author Affiliations & Notes
  • Luisa Sanchez Brea
    Erasmus MC, Rotterdam, Netherlands
  • Danilo Andrade de Jesus
    Erasmus MC, Rotterdam, Netherlands
  • Theo van Walsum
    Erasmus MC, Rotterdam, Netherlands
  • Stefan Klein
    Erasmus MC, Rotterdam, Netherlands
  • Footnotes
    Commercial Relationships   Luisa Sanchez Brea, None; Danilo Andrade de Jesus, None; Theo van Walsum, None; Stefan Klein, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science August 2019, Vol.60, PB0177. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Luisa Sanchez Brea, Danilo Andrade de Jesus, Theo van Walsum, Stefan Klein; Motion correction in 3D-OCT data by intensity-based image registration: an evaluation study. Invest. Ophthalmol. Vis. Sci. 2019;60(11):PB0177.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : A modular open-source registration software (Elastix) is used to systematically study the performance of different registration approaches in 3D-OCT imaging.

Methods : Two public datasets of fovea-centred volumes (6.7x6.7mm) collected with a Bioptigen SD-OCT (NC, USA) were used – 20 Age-related Macular Degeneration (AMD) subjects in dataset 1, and 249 AMD subjects and 115 healthy controls in dataset 2. Each volume consists of 100 B-scans and 1000 A-scans per B-scan. Dataset 1 was used to study inter-observer variability, and dataset 2, for the evaluation of the registration. Intensity-based registration with rigid transformation models (translations (T) and translations plus rotations (E)) applied between consecutive B-scans was considered (Fig. 1). Mean squared difference (MS), normalized correlation (NC), and mutual information (MI) similarity metrics were compared. Manual segmentations of the inner limiting membrane, inner retinal pigment epithelium, and outer Bruch's membrane, by two experts in dataset 1, and by one expert in dataset 2, were used for validation.

Results : Inter-rater variability was calculated as the mean axial difference between the segmentations of two experts on the same B-scan, yielding 4.2±0.7, 7.4±1.4, 5.1±1.1 (μm) for each layer respectively. The average distances between layer segmentations on consecutive B-scans before registration were 16.4±5.6, 17.5±5.4, and 16.5±5.6 for the AMD and 12.2±4.2, 12.3±4.2, and 12.2±4.2 for the control group. All the approaches reduced the misalignments between B-scans observed in the original volumes. The combination of E and MI showed the best results (6.2±3.2, 8.6±3.5, 7.0±3.2 for the AMD group and 4.8±2.3, 5.2±2.4, 5.1±2.4 for the control group). Significant differences (Mann-Whitney U p<0.01) were observed between the AMD and control groups, both in the original volumes and after registration (Fig. 2), and also when comparing the results of different registration approaches in the same group, AMD or Control (Wilcoxon signed-rank p<0.01).

Conclusions : Intensity-based image registration effectively reduces the misalignment between consecutive B-scans. The choice of registration parameters has a significant impact on the accuracy, so further validation and algorithm optimization studies are recommended.

This abstract was presented at the 2019 ARVO Imaging in the Eye Conference, held in Vancouver, Canada, April 26-27, 2019.

 

Registration pipeline.

Registration pipeline.

 

Distribution of the average differences in the unregistered and registered data.

Distribution of the average differences in the unregistered and registered data.

×
×

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.

×