June 2015
Volume 56, Issue 7
ARVO Annual Meeting Abstract  |   June 2015
EyeArt: Automated, High-throughput, Image Analysis for Diabetic Retinopathy Screening
Author Affiliations & Notes
  • Kaushal Solanki
    Eyenuk, Inc., Woodland Hills, CA
  • Chaithanya Ramachandra
    Eyenuk, Inc., Woodland Hills, CA
  • Sandeep Bhat
    Eyenuk, Inc., Woodland Hills, CA
  • Malavika Bhaskaranand
    Eyenuk, Inc., Woodland Hills, CA
  • Muneeswar Gupta Nittala
    Doheny Eye Institute, Los Angeles, CA
  • Srinivas R Sadda
    Doheny Eye Institute, Los Angeles, CA
  • Footnotes
    Commercial Relationships Kaushal Solanki, Eyenuk, Inc (E), Eyenuk, Inc. (P); Chaithanya Ramachandra, Eyenuk, Inc. (E), Eyenuk, Inc. (P); Sandeep Bhat, Eyenuk, Inc. (E), Eyenuk, Inc. (P); Malavika Bhaskaranand, Eyenuk, Inc. (E), Eyenuk, Inc. (P); Muneeswar Nittala, Eyenuk, Inc. (F); Srinivas Sadda, Eyenuk, Inc. (F)
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2015, Vol.56, 1429. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Kaushal Solanki, Chaithanya Ramachandra, Sandeep Bhat, Malavika Bhaskaranand, Muneeswar Gupta Nittala, Srinivas R Sadda; EyeArt: Automated, High-throughput, Image Analysis for Diabetic Retinopathy Screening. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):1429.

      Download citation file:

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

  • Supplements

Fully-automated screening solutions are indispensable to screen the large, ever-increasing number of diabetics for preventable blindness due to diabetic retinopathy (DR). EyeArt meets this need with a computerized, highly accurate, cloud-based DR screening system that enables screening of hundreds of thousands of images in just a few hours. This enables seamless, large-scale screening deployment to aid triage of DR patients in most need of eye-care.


EyeArt system utilizes novel image analysis techniques customized for DR screening and is engineered for large scale deployment on the cloud. The core steps include: (i) image normalization, (ii) non-retinal (lens shot) image rejection, (iii) interest region detection, (iv) multi-scale image description, and (v) advanced machine learning techniques for multi-level classification. EyeArt generates an aggregate DR screening recommendation by analyzing multiple retinal images of a patient.<br /> <br /> EyeArt was evaluated with the large, real-world public dataset (Messidor 2) comprised of images from 874 patients (1748 eyes). The gold standard image grading was obtained from Doheny Eye Institute (DEI) where every image was carefully graded by an expert on the ICDR severity level (0 for no DR and 4 for proliferative DR) and Macular Edema (ME) level (0 for no ME and 1 for ME). The expert used presence of exudates, retinal thickening, and/or microaneurysms, all within 1 disc diameter of fovea, as sign of ME. A patient was deemed non-referable if there was mild or no signs of DR and no ME in both eyes.


EyeArt produced Refer/No Refer screening recommendation for each patient in Messidor2 dataset. EyeArt screening sensitivity was 93.8% (95% CI: 91.0% - 96.6%), specificity of 72.2% (95% CI: 68.6% - 75.8%). This corresponds to 22 false negatives, all of which were of moderate NPDR and did not meet the treatment criteria. No ME cases were missed. The area under the receiver operating characteristic curve (AUROC) was 0.941 (95% CI: 0.920 - 0.959).


The automated screening tool EyeArt has better sensitivity than sensitivities of human graders reported on Messidor2 making it a safe and effective device for DR screening.  

ROC plot for EyeArt in identifying referrable diabetic retinopathy (for the Messidor 2 dataset)
ROC plot for EyeArt in identifying referrable diabetic retinopathy (for the Messidor 2 dataset)


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.