Investigative Ophthalmology & Visual Science Cover Image for Volume 61, Issue 7
June 2020
Volume 61, Issue 7
Free
ARVO Annual Meeting Abstract  |   June 2020
Evaluation of an automatic screening system for detecting retinal pathologies
Author Affiliations & Notes
  • Mathieu Lamard
    LaTIM, Brest University, Brest, France
  • Jean-Bernard Rottier
    Pôle Santé Sud CMCM, France
  • Béatrice Cochener
    LaTIM, Brest University, Brest, France
    CHRU de Brest, France
  • Pascale Massin
    Ophtalmologie, APHP, France
  • Gwenole Quellec
    LaTIM, Inserm, France
  • Sarah Christina Zahida Matta
    LaTIM, Brest University, Brest , France
  • Footnotes
    Commercial Relationships   Mathieu Lamard, None; Jean-Bernard Rottier, None; Béatrice Cochener, None; Pascale Massin, None; Gwenole Quellec, None; Sarah Matta, None
  • Footnotes
    Support  None
Investigative Ophthalmology & Visual Science June 2020, Vol.61, 1640. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Mathieu Lamard, Jean-Bernard Rottier, Béatrice Cochener, Pascale Massin, Gwenole Quellec, Sarah Christina Zahida Matta; Evaluation of an automatic screening system for detecting retinal pathologies. Invest. Ophthalmol. Vis. Sci. 2020;61(7):1640.

      Download citation file:


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

      ×
  • Supplements
Abstract

Purpose : Automatic retinal pathology screening using fundus photograph analysis plays a major role in visual impairment prevention. We have designed an artificial intelligence (AI) capable of automatically differentiating normal photographs from those showing at least one pathological sign. The AI was previously trained on fundus photographs extracted from a Diabetic Retinopathy screening network (OPHDIAT, France). The objective of this study is to evaluate it on a more general population, as part of mass screening in private practice.

Methods : We retrospectively analyzed the data collected in a private screening network in the Le Mans region, France. Fundus images were acquired by orthoptists and read by a single ophthalmologist. Between 2017-2019, 8669 exams were conducted. The mean age of the population was 33 +- 22 years (min: 0, max: 97), and 56% were female. For each examination, only one ground truth annotation was obtained by a senior ophthalmologist. The report included the doctor's opinion: normal, abnormal or of poor quality as well as a free commentary describing, if necessary, the detected pathologies more precisely. AI’s predictions were compared to ground-truth annotations among the 8131 examinations (17120 Images) of sufficient quality.

Results : Compared to human reading, AI is able to differentiate between normal and abnormal examinations with an area under the curve (AUC) of 0.8454. By examining the free comments, we also studied the model's performance by pathology. The AI is particularly efficient at detecting age-related macular degeneration (AUC: 0.9857), degenerative myopia (AUC: 0.9767), but also less frequent pathologies such as coloboma (AUC: 0.9053), epiretinal membranes (AUC: 0.9626) or vascular tortuosities (AUC: 0.9538).

Conclusions : The performance of automatic fundus photograph analysis in a mass screening context is very promising. These performances are probably underestimated considering the ground truth obtained by a single human reader. These results also show the generality of the AI, which was trained on a diabetic population and works very well on a more general population with different interpretation criteria.
Given the severe shortage of ophthalmologists worldwide, this model of automatic retinal pathology screening using fundus photography analysis is therefore a promising solution for earlier patient management.

This is a 2020 ARVO Annual Meeting 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.

×