July 2019
Volume 60, Issue 9
Open Access
ARVO Annual Meeting Abstract  |   July 2019
Optic Disc Tracking in SLO/OCT Videos
Author Affiliations & Notes
  • Chen Gong
    University of Washington, Seattle, Washington, United States
  • John P Kelly
    University of Washington, Seattle, Washington, United States
  • Steven Brunton
    University of Washington, Seattle, Washington, United States
  • Eric Seibel
    University of Washington, Seattle, Washington, United States
  • Footnotes
    Commercial Relationships   Chen Gong, None; John Kelly, None; Steven Brunton, None; Eric Seibel, None
  • Footnotes
    Support  Unrestricted grant from the Peter LeHaye, Barbara Anderson, and William O. Rogers Endowment Funds. ; Department of Mechanical Engineering invention royalty reinvestment program, UW.
Investigative Ophthalmology & Visual Science July 2019, Vol.60, 157. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Chen Gong, John P Kelly, Steven Brunton, Eric Seibel; Optic Disc Tracking in SLO/OCT Videos. Invest. Ophthalmol. Vis. Sci. 2019;60(9):157.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Poor fixation or erroneous eye/head movements cause misaligned optical coherence tomography (OCT) scans and their assessment of retinal nerve fiber layer (RNFL) thickness. The alignment artifact can make centering circular scans on the optic disc impossible. In the case of poor fixation or nystagmus, the line scans near the optic disc are typically acquired in order to get a qualitative assessment of the RNFL. We assessed accurate tracking of the optic disc in SLO/OCT videos from subjects with extreme forms of abnormal eye movements.

Methods : The optic disc template is manually selected from the first frame of the video. For each of the remaining frames, template matching is performed by 1, the current frame is separated into target images which have the same FOV with the template, 2, Principle Component Analysis (PCA) is applied on the target images set to map the target image into a low dimensional space, 3, Map the template into the same space and find its nearest target image. 4. Use SURF feature-point-based registration method to accurately register the template and the selected target image. 5. According to the position of the selected target image, map the template onto the current frame. Using PCA to select the nearest target image provides two similar images for registration. The performance was compared to SIFT and SURF feature detection with RANSAC using OpenCV.

Results : The proposed method is tested in 1860 valid frames of SLO/OCT videos. SURF can find valid feature points in 92% frames after PCA. The root mean square error (RMSE) between the tracked optic disc center and the ground truth is used as the evaluation. The RMSE of the x-axis movement is 2.19 pixels (~0.035 disc diameters), and the RMSE of the y-axis movement is 2.32 pixels (~0.037 disc diameters). The tracking rate is 3 fps. In comparison, SIFT can only locate the disc in < 51% of frames and the error is ~0.2 disc diameters. SURF always failed to detect the disc. Both SIFT and SURF require very large fields for the template to track successfully.

Conclusions : Based on the result, it is validated that our method can track the optic disc accurately in SLO/OCT videos and indicate the deformation of the object. The failure cases in several low contrast frames do not influence the tracking quality of the following frames. Using this method, we can acquire the line scans near the optic disc for RNFL especially from people with poor fixation.

This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.

×
×

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.

×