June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Transfer Learning reveals differences in arterio-venous oxygenation biomarkers in patients with glaucoma and healthy controls
Author Affiliations & Notes
  • Lucas William Rowe
    Indiana University School of Medicine, Indianapolis, Indiana, United States
  • Alon Harris
    Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Giovanna Guidoboni
    University of Missouri, Columbia, Missouri, United States
  • Alice Chandra Verticchio Vercellin
    Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Aaron Beckwith
    University of Missouri, Columbia, Missouri, United States
  • Rajat Rai
    University of Missouri, Columbia, Missouri, United States
  • James Keller
    University of Missouri, Columbia, Missouri, United States
  • Christopher Wikle
    University of Missouri, Columbia, Missouri, United States
  • Erin L. Robinson
    University of Missouri, Columbia, Missouri, United States
  • Maggie Lin
    University of Missouri, Columbia, Missouri, United States
  • Daphne Zou
    University of Missouri, Columbia, Missouri, United States
  • Amber Wolf
    Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Roberto Nunez
    University of Missouri, Columbia, Missouri, United States
  • Danny Kim
    Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Brent A Siesky
    Icahn School of Medicine at Mount Sinai, New York, New York, United States
  • Footnotes
    Commercial Relationships   Lucas Rowe None; Alon Harris AdOM, Qlaris, Luseed, Cipla , Code C (Consultant/Contractor), AdOM, Luseed, Oxymap, Qlaris, Phileas Pharma, SlitLed, QuLent, Code I (Personal Financial Interest), AdOM, Qlaris, Phileas Pharma, Code S (non-remunerative); Giovanna Guidoboni Foresite Healthcare LLC, Code C (Consultant/Contractor), Gspace LLC, Code I (Personal Financial Interest); Alice Chandra Verticchio Vercellin None; Aaron Beckwith None; Rajat Rai None; James Keller None; Christopher Wikle None; Erin Robinson None; Maggie Lin None; Daphne Zou None; Amber Wolf None; Roberto Nunez None; Danny Kim None; Brent Siesky None
  • Footnotes
    Support  The work was partially supported by the NSF awards DMS 1853222/2021192 and DMS 2108711/2108665, NYEE Foundation grant, and in part by a Challenge Grant award from Research to Prevent Blindness, NY.
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2019 – A0460. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Lucas William Rowe, Alon Harris, Giovanna Guidoboni, Alice Chandra Verticchio Vercellin, Aaron Beckwith, Rajat Rai, James Keller, Christopher Wikle, Erin L. Robinson, Maggie Lin, Daphne Zou, Amber Wolf, Roberto Nunez, Danny Kim, Brent A Siesky; Transfer Learning reveals differences in arterio-venous oxygenation biomarkers in patients with glaucoma and healthy controls. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2019 – A0460.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Lower retinal arterio-venous oxygen saturation (A/V-diff) has been reported in patients with open angle glaucoma (OAG). However, utilization of biomarkers of reducted oxygen extraction in retinal tissues is complex. This analysis applies Transfer Learning (TL) of glaucoma progression data outcomes to analyze a prospective cross-sectional sample of OAG and healthy controls retinal photographic oximetry images.

Methods : Fuzzy c-Means (FCM) clustering is applied to a prospective glaucoma progression dataset (n=115) enhanced with hemodynamic variables predicted by a validated mathematical model (Guidoboni et al, IOVS, 2014). The model specifically uses intraocular pressure (IOP), blood pressure (BP) and heart rate (HR) as individualized model inputs. The 3 FCM-based clusters are then applied to categorize cross-sectional data (n=45 healthy, 11 OAG) based on subjects mean arterial pressure (MAP) and IOP. Optical coherence tomography (OCT) biomarkers of retinal nerve fiber layer thickness (RNFL) and the optic nerve head and oximetry biomarkers were analyzed by (i) distinguishing between healthy and OAG subjects and then (ii) sectioning each group by the 3 FCM-based clusters.

Results : Table 1 displays the FCM TL data for patients with OAG and healthy controls. Structural markers are worse in OAG than healthy eyes in both analyses (i) and (ii). Superior temporal (ST) A/V-diff reveal a 7% increase from healthy to OAG eyes (p=0.12) while inferior temporal (IT) A/V-diff display a 15% increase (p=0.07). When further sectioned, there was an increase in the ST A/V-diff in cluster 1 by 12% (p=0.21), but in cluster 3 there was no change (p=0.63); IT A/V-diff shows a 10% decrease (p=0.46) in cluster 1, but a 9% increase (p=0.09) in cluster 3 (p values via 2-sample Wilcoxon signed rank test for medians).

Conclusions : Specific clustering from TL confirms worse OCT structural markers for OAG eyes in all clusters. It further reveals A/V differences among clusters that were masked when subgrouping was not considered, with no change in cluster 3 for ST A/V-diff, and a decrease in cluster 1 for IT A/V-diff. TL and FCM clustering of clinical inputs may improve specificity of individualized risk assessment.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Table 1. Analysis of cross-sectional data without FCM-based clustering and analysis of cross-sectional data with FCM-based clustering

Table 1. Analysis of cross-sectional data without FCM-based clustering and analysis of cross-sectional data with FCM-based clustering

×
×

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.

×