June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Convolutional Neural Network for Glaucoma detection using Compass color fundus images
Author Affiliations & Notes
  • CHIARA RUI
    Centervue, Padova, Italy
  • Silvia Gazzina
    Centervue, Padova, Italy
  • Giovanni Montesano
    Optometry and Visual Sciences, City, University of London, London, United Kingdom
  • David P. Crabb
    Optometry and Visual Sciences, City, University of London, London, United Kingdom
  • David F Garway-Heath
    NIHR Biomedical Research Centre, Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology, London, United Kingdom
  • Francesco Oddone
    Glaucoma Unit, IRCCS GB Bietti Eye Foundation, Roma, Italy
  • Paolo Lanzetta
    Department of Ophthalmology, Universita degli Studi di Udine, Udine, Italy
  • Paolo Brusini
    Department of Ophthalmology, “Città di Udine” Health Center, Udine, Italy
  • Chris A Johnson
    Department of Ophthalmology and Visual Sciences, University of Iowa Hospitals and Clinics, Iowa, United States
  • Paolo Fogagnolo
    Eye Clinic, Universita degli Studi di Milano, Milano, Italy
  • Luca M Rossetti
    Eye Clinic, Universita degli Studi di Milano, Milano, Italy
  • Footnotes
    Commercial Relationships   CHIARA RUI Centervue , Code E (Employment); Silvia Gazzina Centervue , Code E (Employment); Giovanni Montesano Centervue , Code C (Consultant/Contractor); David Crabb None; David Garway-Heath Centervue , Code C (Consultant/Contractor); Francesco Oddone Centervue , Code C (Consultant/Contractor); Paolo Lanzetta Centervue , Code C (Consultant/Contractor); Paolo Brusini None; Chris Johnson None; Paolo Fogagnolo Centervue , Code C (Consultant/Contractor); Luca Rossetti Centervue , Code C (Consultant/Contractor)
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2039 – A0480. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      CHIARA RUI, Silvia Gazzina, Giovanni Montesano, David P. Crabb, David F Garway-Heath, Francesco Oddone, Paolo Lanzetta, Paolo Brusini, Chris A Johnson, Paolo Fogagnolo, Luca M Rossetti; Convolutional Neural Network for Glaucoma detection using Compass color fundus images. Invest. Ophthalmol. Vis. Sci. 2022;63(7):2039 – A0480.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : To compare the performance of a Convolutional Neural Network (CNN) model with experienced glaucoma specialists’ grading in detecting glaucomatous optic neuropathy using color images collected with the Compass fundus perimeter (CMP, Centervue, Italy).

Methods : The data used for this project were collected during an international multicentric clinical study. The dataset consisted of 1930 color images of the Optic Nerve Head (ONH, 400x400 pixels) automatically captured by the Compass. Each image was labelled as healthy (NRM) or glaucoma (GLC) based on a glaucoma experts’ evaluation (clinical judgment of the nerve including ophthalmoscopy, fundus image and OCT) performed during the main study, for a total of 1010 NRM and 920 GLC.The original image was resized to 200x200 pixels for the CNN. The CNN model consisted of sequential convolutional and max pooling layers for a total of 1,278,113 trainable parameters. The model was trained using 90% of the dataset (80% train, 10% validation) and tested using the remaining 10% of the data randomly sampled across all study sites. The output of the final sigmoid activation function (AF) was used to build a Receiver Operating Characteristic (ROC) curve. Additionally, two experienced glaucoma specialists (clinician 1 and 2), involved in the main study, were asked to independently grade the test set images with no extra information and no knowledge of group assignment. Their performance was compared to that of the CNN at the same specificity on the ROC curve.

Results : The performance obtained by the CNN, clinician 1 and clinician 2 on the test set (101 NRM, 92 GLC) is shown in Figure 1 and reported in Table 1. The sensitivity of Clinician 2 was just below the 2.5%-Confidence bound of the CNN. The table also reports the performance of the CNN for a threshold of 0.5 on the AF.

Conclusions : The CNN had a performance that ranged from similar to significantly better than the expert graders. The methodology could have applications for automated case finding in glaucoma care.

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

 

Figure 1. CNN ROC curve and 95% Confidence Intervals with dots representing clinician 1, clinician 2 and CNN performance.

Figure 1. CNN ROC curve and 95% Confidence Intervals with dots representing clinician 1, clinician 2 and CNN performance.

 

Table 1. Performance of clinician 1, clinician 2 and CNN. For each specificity value obtained by clinician 1 and 2, the corresponding CNN sensitivity with 95% Confidence Intervals Bounds were reported.

Table 1. Performance of clinician 1, clinician 2 and CNN. For each specificity value obtained by clinician 1 and 2, the corresponding CNN sensitivity with 95% Confidence Intervals Bounds were reported.

×
×

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.

×