Investigative Ophthalmology & Visual Science Cover Image for Volume 61, Issue 9
July 2020
Volume 61, Issue 9
Free
ARVO Imaging in the Eye Conference Abstract  |   July 2020
Change in geographic atrophy lesion area: comparison between fully automatic segmentation and semi-automatic segmentation
Author Affiliations & Notes
  • Jasmine Patil
    Genentech, South San Francisco, California, United States
  • Neha Anegondi
    Genentech, South San Francisco, California, United States
  • Verena Steffen
    Genentech, South San Francisco, California, United States
  • Simon Gao
    Genentech, South San Francisco, California, United States
  • Footnotes
    Commercial Relationships   Jasmine Patil, Genentech (E); Neha Anegondi, Genentech (E); Verena Steffen, genentech (E); Simon Gao, Genentech (E)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science July 2020, Vol.61, PB0062. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jasmine Patil, Neha Anegondi, Verena Steffen, Simon Gao; Change in geographic atrophy lesion area: comparison between fully automatic segmentation and semi-automatic segmentation . Invest. Ophthalmol. Vis. Sci. 2020;61(9):PB0062.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Change in geographic atrophy(GA) lesion area from fundus autofluorescence(FAF) images is an approvable clinical trial endpoint. This endpoint is computed using fully automatic segmentation, and its performance is compared against semi-automatic segmentation.

Methods : This retrospective study was conducted using data from patients with GA in ProximaA(NCT02479386), ProximaB(NCT02399072) and Exposure(NCT02288559) clinical trials. The automatic segmentation network Unet 1024 was trained on 400 FAF images(96 patients) with the corresponding semi-automatically segmented GA lesion masks from the Exposure study. The model performance was tested with 141 images(82 patients) from ProximaA and 322 images(175 patients) ProximaB. We calculated the mean dice score between the semi-automatic annotation and automatic segmentation. To analyze the performance of automatic segmentation with respect to the clinical endpoint of change in the GA lesion area, we compared the GA area change derived from semi-automatic segmentation and automated segmentation. The GA area change(absolute change in mm2) was calculated between month6 & screening and month12 & screening.

Results : The mean dice score between the semi-automatic and automatic segmentation was 0.90(σ=0.11) for ProximaA and 0.9(σ=0.11) for ProximaB. The mean dice score between two readers’ semi-automatic segmentation was 0.95(σ=0.08) in both ProximaA and ProximaB studies. The R2 between two readers’ longitudinal GA lesion area(mm2) change was 0.95 at month 6 & screening and 0.97 at month12 & screening in ProximaA and ProximaB studies. Fig1 shows that the R2 for the GA lesion area change for ProximaA was 0.26 between month6 and screening(a) and 0.64 between month12 and screening(b). For ProximaB the R2 is 0.25 between month6 and screening(c) and 0.60 between month12 and screening(d).

Conclusions : In comparison with the semi-automatic segmentation of GA lesions, fully automatic segmentation performance was acceptable with respect to the dice score. However, results demonstrate that the change in the GA lesion area(mm2) between semi-automatic and automatic segmentation results had a low correlation. The correlation improved over a longer time interval; the correlation between month12 to screening increased from that of month6 to screening. Further investigation is required to understand the variance caused by the model segmentation.

This is a 2020 Imaging in the Eye Conference abstract.

 

×
×

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.

×