June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Fully automated Deep Learning analysis of hemoglobin images of the optic nerve associated with perimetry for the study of glaucoma
Author Affiliations & Notes
  • Manuel Gonzalez de la Rosa
    Instrumentacion y Oftalmologia INSOFT SL, Spain
  • Daniel Gonzalez-Hernandez
    Instrumentacion y Oftalmologia INSOFT SL, Spain
  • Marta Gonzalez-Hernandez
    Ophthalmology, Hospital Universitario de Canarias, La Laguna, Canarias, Spain
  • Daniel Perez-Barbudo
    Instrumentacion y Oftalmologia INSOFT SL, Spain
  • Footnotes
    Commercial Relationships   Manuel Gonzalez de la Rosa, INSOFT SL (I), INSOFT SL (C), INSOFT SL (P); Daniel Gonzalez-Hernandez, INSOFT SL (I), INSOFT SL (E), INSOFT SL (P); Marta Gonzalez-Hernandez, INSOFT SL (I), INSOFT SL (E), INSOFT SL (P); Daniel Perez-Barbudo, INSOFT SL (E)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 1009. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Manuel Gonzalez de la Rosa, Daniel Gonzalez-Hernandez, Marta Gonzalez-Hernandez, Daniel Perez-Barbudo; Fully automated Deep Learning analysis of hemoglobin images of the optic nerve associated with perimetry for the study of glaucoma. Invest. Ophthalmol. Vis. Sci. 2021;62(8):1009.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Colorimetric analysis of optic nerve images for assessing their hemoglobin distribution (Laguna ONhE) (1-4) is tested in combination with perimetry.

Methods : Deep learning training was used to identify nerve edges, laterality of the eye, image quality, vessel segmentation and classification (normal vs glaucoma). Data was compiled into a "Globin Distribution Function" (GDF), which was also associated with visual field irregularity indices: Pattern Standard Deviation (PSD), square root of loss variance (sLV), and threshold coefficient of variation (TCV) (5).
477 normal eyes and 333 confirmed and suspected glaucoma eyes, which were examined with three fundus cameras, two perimeters and two visual field strategies. The results were compared with Cirrus OCT.

Results : GDF sensitivity identifying glaucoma was 75.7% for a specificity of 99.0%. The most sensitive OCT index was the Rim Area (sensitivity 67.0%, P=0.0131). Its association with visual field irregularity produced the following AUC's: GDF&PSD-sLV = 0.963-0.986 and GDF&TCV = 0.965-0.987, while Rim Area&PSD = 0.927-0.960, Vertical Cup/Disc&PSD = 0.929-0.961 and RNFLT&PSD = 0.894-0.933 (P<0.0001 in all cases). For 99% specificity, GDF&TCV achieved 80.8% sensitivity and RNFLT&PSD 72.4%.
In cases where the morphological or functional indices had an unusual level in regard to 95% of normal subjects, the GDF&TCV achieved AUC's of 0.99-1.00 and sensitivities of 87.3-96.0% for 99% specificity.

Conclusions : Laguna ONhE associated to perimetry offers relevant diagnostic results in glaucoma, although new studies might be necessary to consolidate such results.

References:
1- Gonzalez de la Rosa M et al. IOVS 2013;54:482-489.
2- Gonzalez-Hernandez M et al. J Ophthalmol 2017;2340236.
3- Gonzalez-Hernandez D et al. J Clin Exp Opthamol2018: 9:5 doi:10.4172/2155-9570.1000760.
4- Mendez-Hernandez C et al. Br. J. Ophthalmol 2020;doi:10.1136/bjophthalmol-2020-316455.
5- Abreu-Gonzalez R et al. Eur J Ophthalmol 2018;28:590-597.

This is a 2021 ARVO Annual Meeting abstract.

 

Fig. 1. ROC curves obtained with Laguna ONhE, OCT-Cirrus and perimetry indices and sensitivities for 95%-99% specificities.

Fig. 1. ROC curves obtained with Laguna ONhE, OCT-Cirrus and perimetry indices and sensitivities for 95%-99% specificities.

 

Fig.2. ROC curves of the combinations between Laguna ONhE and OCT-Cirrus with perimetry.

Fig.2. ROC curves of the combinations between Laguna ONhE and OCT-Cirrus with perimetry.

×
×

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.

×