June 2020
Volume 61, Issue 7
ARVO Annual Meeting Abstract  |   June 2020
Using computer vision to overcome nystagmus in OCT
Author Affiliations & Notes
  • John Hensel
    Massachusetts Eye and Ear, Boston, Massachusetts, United States
  • Footnotes
    Commercial Relationships   John Hensel, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2020, Vol.61, 3248. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      John Hensel; Using computer vision to overcome nystagmus in OCT. Invest. Ophthalmol. Vis. Sci. 2020;61(7):3248.

      Download citation file:

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

  • Supplements

Purpose : Patients with severe nystagmus present a unique challenge to optical coherence tomography (OCT) imaging. When a rapidly moving eye cannot be tracked, the scan is difficult to reliably position over the fovea, the resulting image does not resolve fine structural details, and comparing between visits is not directly possible. These limitations make it difficult to track fine details in disease progression as well as responses to potential therapies. This paper presents a novel method utilizing a unique scanning protocol and computer vision (CV) processing to generate high-quality and repeatable single-line scans.

Methods : Two patients with severe nystagmus due to advanced retinitis pigmentosa (RP) and achromatopsia (ACHM), were scanned as part of normal clinical care. A series of high resolution volume scans with an 11 micron line separation was performed on a Heidelberg Spectralis with tracking disabled. Due to the movement of the eye with and against the advance of the scan line, each volume scan contained multiple b-scans over the fovea. The volume scans were exported as individual b-scans for processing.

After the operator identified a high quality scan over the fovea (the “key”) from among the exported images, a program run in Fiji / ImageJ processed the scans. The program ran SIFT (scale invariant feature transform) to identify a set of unique “features” of the key from which it could match every subsequent image. The images that met the match criteria were given a ridged transform to register to the key image. A final synthetic scan was then generated by taking the median value at each pixel for the matched images.

For one patient, subsequent encounters were scanned using the same parameters. The program then used the previously specified key to first find a best match from the new set of images and then seed it as the new key thus allowing to image a specified location over time.

Results : We demonstrate that this method can be used successfully on patients with severe nystagmus. Scanning, exporting, and processing had a typical time of 19 minutes per eye with minimal operator intervention.

Conclusions : This is a promising method for scanning of challenging patients. Further research to evaluate accuracy is needed.

This is a 2020 ARVO Annual Meeting abstract.


ACHM patient, using method to generate scans at the same location at different visits

ACHM patient, using method to generate scans at the same location at different visits


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.