June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Deep learning (DL)-based morphometric analysis of retinal ganglion cell (RGC) axons in a rat model of glaucoma
Author Affiliations & Notes
  • C Ross Ethier
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
    Biomedical Engineering, Emory University School of Medicine, Atlanta, Georgia, United States
  • Vidisha Goyal
    Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Gabriela Sánchez-Rodríguez
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Dillon Brown
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Bailey Hannon
    Mechanical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Aaron M. Toporek
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Matthew D. Ritch
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Andrew Feola
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
    Atlanta VA Medical Center, Decatur, Georgia, United States
  • Arthur Thomas Read
    Biomedical Engineering, Georgia Institute of Technology, Atlanta, Georgia, United States
  • Footnotes
    Commercial Relationships   C Ethier, None; Vidisha Goyal, None; Gabriela Sánchez-Rodríguez, None; Dillon Brown, None; Bailey Hannon, None; Aaron Toporek, None; Matthew Ritch, None; Andrew Feola, None; Arthur Read, None
  • Footnotes
    Support  NIH R01 EY025286 (CRE), 5T32 EY007092-32 (BGH), Department of Veteran Affairs R&D Service Career Development Award (RX002342; AJF), and Georgia Research Alliance (CRE).
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 2346. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      C Ross Ethier, Vidisha Goyal, Gabriela Sánchez-Rodríguez, Dillon Brown, Bailey Hannon, Aaron M. Toporek, Matthew D. Ritch, Andrew Feola, Arthur Thomas Read; Deep learning (DL)-based morphometric analysis of retinal ganglion cell (RGC) axons in a rat model of glaucoma. Invest. Ophthalmol. Vis. Sci. 2021;62(8):2346.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : We developed a DL algorithm to identify axoplasm (‘Ax’) and myelin sheaths of normal-appearing RGC axons from light micrographs of optic nerve (ON) cross-sections [Goyal+, submitted ARVO 2021]. Here we: (i) compare RGC Ax area (‘Area’) and eccentricity (‘Ecc’) between manually- and DL-segmented images; and (ii) compare axon density (axons/μm2) and axon size distributions between hypertensive and control eyes.

Methods : Ocular hypertension was induced unilaterally in 14 Brown-Norway rats (3-13 months; 12 male:2 female) by episcleral hypertonic saline or microbead injection. Contralateral eyes were controls. (i) To compare manual and DL segmentations, we extracted the mean and median Ax Area and Ecc from all images, and computed the mean absolute percentage error (MAPE) between them for all ON. (ii) To compare control vs. hypertensive eyes, we restricted analysis to pairs where the hypertensive eye had an IOP burden > 100 mmHg day (n=7 pairs). Axons were counted in ON sub-images, divided by sub-image area to obtain axon density, and compared hypertensive vs. control. Axons were grouped by Area into 9 bins, and for each ON and each bin the ratio (hypertensive axon density)/(control axon density) was computed and normalized following Quigley+ [PMID: 3583630].

Results : (i) Mean and median Area agreed well between manual and DL segmentations (MAPE < 4.6%, Fig 1), as did Ecc (MAPE < 1.41%). (ii) Hypertensive eyes showed lower axon density (0.19 ± 0.11 vs. 0.36 ± 0.08 axons/μm2), an approximate 40% decrease (p= 0.038, 2-sided paired t-test). We did not observe a preferential loss of large axons with hypertension (Fig 2).

Conclusions : Morphometric parameters of RGC axons extracted by a DL algorithm agreed well with manual segmentations. The algorithm detected decreased axon density in hypertensive eyes, but no preferential loss of large axons (perhaps due to the rat model used). This algorithm provides a fast and non-subjective method to quantify axonal loss in animal models of glaucoma; future consideration of axonal shape changes may add sensitivity to detection of glaucomatous damage.

This is a 2021 ARVO Annual Meeting abstract.

 

Comparison between DL- and manually-segmented axoplasm area and eccentricity in control and hypertensive eyes.

Comparison between DL- and manually-segmented axoplasm area and eccentricity in control and hypertensive eyes.

 

Normalized ratio of axon density (hypertensive/control) vs. Ax Area. No preferential loss of large axons is seen. The largest bin contained all Ax with Area ≥ 2.69 μm2.

Normalized ratio of axon density (hypertensive/control) vs. Ax Area. No preferential loss of large axons is seen. The largest bin contained all Ax with Area ≥ 2.69 μm2.

×
×

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.

×