June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Predicting ab interno trabeculotomy (Trabectome) surgical outcomes using pre-op sclera images from IOL Master 700: a retrospective study utilizing deep learning
Author Affiliations & Notes
  • Melissa Chang
    Ophthalmology, University of California Irvine, Irvine, California, United States
  • Anjali Herekar
    Ophthalmology, University of California Irvine, Irvine, California, United States
  • Junhan Ouyang
    University of California Irvine, Irvine, California, United States
  • Ken Y. Lin
    Ophthalmology, University of California Irvine, Irvine, California, United States
  • Footnotes
    Commercial Relationships   Melissa Chang None; Anjali Herekar None; Junhan Ouyang None; Ken Lin None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2023, Vol.64, OD43. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Melissa Chang, Anjali Herekar, Junhan Ouyang, Ken Y. Lin; Predicting ab interno trabeculotomy (Trabectome) surgical outcomes using pre-op sclera images from IOL Master 700: a retrospective study utilizing deep learning. Invest. Ophthalmol. Vis. Sci. 2023;64(8):OD43.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Minimally invasive glaucoma surgery (MIGS) has become increasingly popular despite variable surgical success. One theory of prognostic factors includes characteristics of the episcleral vessels, which provide insights into the state of aqueous outflow. This study applies deep learning to routine preoperative scleral images from IOL Master 700 to predict MIGS outcomes in patients undergoing combined cataract surgery and trabeculotomy ab externo (Trabectome).

Methods : 163 unique adult eyes that underwent cataract extraction with intraocular lens (IOL) placement and Trabectome between January 2018 - March 2022 at our tertiary medical center were included. The average age of participants was 71.1±9.0 years. There were 98 males and 62 females.
Success of MIGS was defined as a decrease in the number of glaucoma drops and/or a 20% reduction in IOP by three months post surgery.
We tested several deep learning models. The validation accuracies were evaluated using stratified K-fold. The reported mean and variances were based on 15 runs. Left eyes were flipped horizontally to align nasal and temporal regions across all scleral images.

Results : At baseline, 74.23% of procedures were successful at 3 months postoperatively.
The best performing model yielded the highest validation accuracy of 78.04 ± 1.27 and a false negative rate of 1.59 ± 2.86.
We subdivided each scleral image into a 3-by-3 sub-areas, with the middle area corresponding to the pupil. Obscuring the entire image and then selectively unblocking each sub-area at a time (except for the lower middle area) significantly increased validation accuracy compared to unblocking the middle sub-area (p<0.01). This suggested that our algorithm was relying on limbus-containing zones to formulate its prediction.

Conclusions : We successfully utilized deep learning to predict 3-month post-op success of combined trabectome and cataract surgery above baseline surgical success of our dataset. The results from the regional analysis of the images suggest that the deep learning model utilizes predominantly the scleral regions of the eye to predict surgical outcome, as opposed to the middle pupil. The non-invasive scleral imaging technique using the IOL Master 700 is a novel tool to help identify optimal surgical candidates and ultimately improve surgical outcomes for a variety of ocular procedures.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

×
×

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.

×