June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
External Validation of Autonomous Retinopathy of Prematurity Screening in the SUNDROP Program
Author Affiliations & Notes
  • Tom Murickan
    National Library of Medicine, Bethesda, Maryland, United States
  • Aaron S Coyner
    Oregon Health & Science University, Portland, Oregon, United States
  • Susan Ostmo
    Oregon Health & Science University, Portland, Oregon, United States
  • Darius M Moshfeghi
    Stanford University School of Medicine, Stanford, California, United States
  • Robison Vernon Paul Chan
    University of Illinois Chicago College of Medicine, Chicago, Illinois, United States
  • Jayashree Kalpathy-Cramer
    Athinoula A Martinos Center for Biomedical Imaging, Charlestown, Massachusetts, United States
  • Michael F Chiang
    National Library of Medicine, Bethesda, Maryland, United States
    National Eye Institute, Bethesda, Maryland, United States
  • J. Peter Campbell
    Oregon Health & Science University, Portland, Oregon, United States
  • Footnotes
    Commercial Relationships   Tom Murickan None; Aaron Coyner None; Susan Ostmo None; Darius Moshfeghi None; Robison Chan None; Jayashree Kalpathy-Cramer None; Michael Chiang None; J. Peter Campbell None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 268. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Tom Murickan, Aaron S Coyner, Susan Ostmo, Darius M Moshfeghi, Robison Vernon Paul Chan, Jayashree Kalpathy-Cramer, Michael F Chiang, J. Peter Campbell; External Validation of Autonomous Retinopathy of Prematurity Screening in the SUNDROP Program. Invest. Ophthalmol. Vis. Sci. 2023;64(8):268.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Retinopathy of prematurity (ROP) screening is effective for secondary prevention of blindness from ROP, but several regions lack sufficient resources to screen all at-risk babies. Previous work suggests that artificial intelligence algorithms, such as the i-ROP DL system, can be used for quantitative diagnosis of ROP through the assignment of a vascular severity score (VSS). This would be useful for both assistive diagnosis and autonomous ROP screening. In this study, we evaluated various cutoffs for effectiveness of ROP screening on an external ROP telemedicine dataset.

Methods : The SUNDROP dataset was collected from 14 hospitals between 01/2013 and 12/2021. A total of 79807 images were collected from 1570 patients over 6310 exams. All eyes were graded by a single expert grader, assigned a diagnosis according to the International Classification of ROP (ICROP) and assigned a category of no/mild ROP, “more than mild ROP” (MTMROP) and TR-ROP for evaluation. The i-ROP DL system was used to apply a VSS to each eye exam using methods previously published. We compared the VSS output by ICROP classification, and calculated the area under the receiver operating characteristic curve (AUROC) for detection of MTMROP and TR-ROP using the i-ROP DL system.

Results : Figure 1 shows the VSS by category, and zone and stage (P<0.05 for all comparisons), and Figure 2 shows the AUROC curve for MTMROP (0.90) and TR-ROP (0.99). A cutoff of >2.0 yielded 82% sensitivity and 81% specificity for MTMROP detection and 100% sensitivity and 78% specificity for TR-ROP detection. A cutoff of >3.0 yielded 100% sensitivity and 94% specificity for TR-ROP detection, but sensitivity for MTMROP detection dropped to 52%. Theoretically, at cutoffs 2.0 and 3.0, 72.1% and 93.7%, respectively, of exams could have been read autonomously without missing any cases of TR-ROP.

Conclusions : These results suggest that autonomous ROP screening using a VSS may be feasible, with moderate sensitivity and specificity for detection of MTMROP at a referral cutoff of 2.0 or greater. In regions with insufficient resources for ophthalmoscopic ROP screening, autonomous ROP screening could provide rapid diagnosis of MTMROP with low chance of missing TR-ROP.

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

 

Fig. 1: Boxplots showing (A) 3 categories of ROP severity and associated VSS and (B) VSS for each zone, subdivided by ROP stage.

Fig. 1: Boxplots showing (A) 3 categories of ROP severity and associated VSS and (B) VSS for each zone, subdivided by ROP stage.

 

Fig. 2: ROC curves for i-ROP DL detection of MTMROP and TR-ROP.

Fig. 2: ROC curves for i-ROP DL detection of MTMROP and TR-ROP.

×
×

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.

×