Menu

  1. Home
  2. /
  3. Community / Social Ophthalmology–I
  4. /
  5. FP1296 : ON-SITE EVALUATION...

Share this post

Dr. Payal Shah, Dr. Mahesh Shanmugam P, Dr. Mishra Divyansh Kailashcandra, Dr. Rajesh Ramanjulu

Introduction:

India has one of the highest prevalence of diabetes in the world with over 72 million diabetics and prevalence of around 8.8%.1The prevalence of diabetic retinopathy(DR) in Indian diabetics has been reported to be around 182-27%.3 This emphasizes the need for proper diabetic screening to reduce the burden of diabetic retinopathy related blindness. However, paucity of trained retinal specialists in India limits effective screening of asymptomatic patients, thereby resulting in patients presenting late with advanced diabetic eye disease.

Fundus photograph based diabetic retinopathy screening in lieu of physical screening can be performed using manual grading of fundus images by trained graders or retina specialists. The technology of machine based learning to detect DR has given a new horizon for DR screening and is improving rapidly. Its application in diagnosing referable DR patients would have a great impact in reducing the blindness burden. Several studies have shown that a 500 posterior pole fundus image alone can be used as a screening tool to identify DR.4,5

Aim:

To evaluate the on-site efficacy of artificial intelligence(AI) based diabetic retinopathy(DR) detection in the setting of tele-DR screening.

Materials and methods:

This was a prospective study carried out at various eye screening camps in semi-urban and rural areas around Bangalore. 50 degree posterior pole macula centredundilated and dilated fundus images of diabetic patients were captured by trained optometrists using Trinetra non-mydriatic fundus camera (Forus, India).

Using Google forms, we prepared a questionnaire which consisted of patient’s epidemiologic data, diabetic status, visual acuity, lens status, undilated and dilated fundus photographs. This app was integrated with Google spread sheet and converted into a mobile application by using appsheet.

Netra.AI is an artificial intelligence(AI) based machine learning algorithm developed to detect diabetic retinopathy by Lebencare technologies. We used this algorithm to identify DR in our study.

The optometrists entered the data using mobile app and uploaded the fundus images onto the mobile application from the camp site. The images were graded by a retina specialist at the base hospital and simultaneously by Netra.AI for automated DR detection.

Referable DR(RDR)/Prompt Referral was defined as moderate NPDR or above and/or presence of macular edema.

Results:

989 diabetic patients were screened. Of the 1978 eyes, 1540 eyes were evaluated for DR. Remaining eyes were excluded due to dense cataract.

Of the 1540 eyes, 459 eyes had No DR, 1081 eyes were identified to have any diabetic retinopathy, which included 220 eyes with Mild NPDR, 542 eyes with moderate NPDR, 138 eyes with severe NPDR and 181 with PDR according to the manual grading. Overall, 861 eyes were identified as referable DR.

The AI based grading showed446 eyes with No DR, 1094 eyes with Any DR and 852 eyes were correctly identified as referable DR. Overall sensitivity to identify any DR was 98.5% and specificity was 93.5%.

Overall sensitivity of AI to identify referable DR was 99% and specificity was 98.3%.

Discussion:

Simple utilisation of Google forms can help us to develop a mobile application by ourselves requiring no specialised software coding. This can have a great utility in telescreening programs at minimal cost.

In the present scenario, every diabetic patient needs to be referred to a retina specialist for diagnosis and treatment of DR. If an automated software is able to identify sight threatening DR as precisely as a clinician, it would greatly reduce the burden of screening on vitreoretinal(VR) surgeons.

AI has been shown to have a sensitivity of more than 90% to detect referable DR in the studies so far performed. The sensitivity and specificity of AI improves as the machine sees more and more images. It would also help to reduce the financial burden on the patient. There are limited studies6-10 published in literature on use of deep convolutional neural networks for automated DR detection, summary of which is shown in Table below.

AI can help not only in strengthening the screening services with a greater reach and fulfilling the resource gap, but also in reducing the turnaround time and costs. Netra.AI takes an average of 5 seconds per image for detecting DR and marking lesions, which is as good as a clinician.  The advantage of AI would thus be in its ability to detect DR without the need for a trained retina specialist, remote screening, ability to rapidly screen large numbers. This can revolutionise telescreening in ophthalmology, especially where people don’t have access to specialised health care. Integrated into fundus cameras, an optometrist or a trained technician can screen diabetic patients much earlier using portable non-mydriatic cameras.

References:

  1. International Diabetes Federation. IDF Diabetes Atlas – 7th edition. International Diabetes Federation. 2015. Accessed July 2017.
  2. Raman R, Rani PK, ReddiRachepalle S, et al. Prevalence of diabetic retinopathy in India: SankaraNethralaya Diabetic Retinopathy Epidemiology and Molecular Genetics Study report 2. Ophthalmology. 2009;116(2):311-318.
  3. Shah S, Das AK, Kumar A, Unnikrishnan AG, Kalra S, Baruah MP, et al. Baseline characteristics of the Indian cohort from the IMPROVE study: A multinational, observational study of biphasic insulin aspart 30 treatment for type 2 diabetes. AdvTher2009;26:325-35.
  4. Bawankar P, Shanbhag N, et al Sensitivity and specificity of automated analysis of single-field non-mydriatic fundus photographs by Bosch DR Algorithm-Comparison with mydriatic fundus photography (ETDRS) for screening in undiagnosed diabetic retinopathy. PLoS One. 2017 Dec 27;12(12):e0189854.
  5. Srihatrai P, Hlowchitsieng T. The diagnostic accuracy of single- and five-field fundus photography in diabetic retinopathy screening by primary care physicians. Indian J Ophthalmol. 2018 Jan;66(1):94-97.
  6. Gargeya R, Leng T. Automated identification of diabetic retinopathy using deep learning. Ophthalmology. 2017;124(7):962-969.
  7. Pratt H, Coenen F, Broadbent DM, Harding SP, Zheng Y. Convolutional neural networks for diabetic retinopathy.ProcediaComputSci 2016;90(1):200–5.
  8. Gulshan V, Peng L, Coram M, et al. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. 2016;304:649-656.
  9. Wong TY, Bressler NM. Artificial intelligence with deep learning technology looks into diabetic retinopathy screening. JAMA. 2016;316(22):2366-2367
  10. Ramachandran N, Hong SC, Sime MJ, Wilson GA. Diabetic retinopathy screening using deep neural network. ClinExpOphthalmol. 2018 May;46(4):412-416.
Dataset Sensitivity Specificity
Any DR
Pratt et al7

(U.K)

 

 5000 images 30% 95%
Gargeya et al6

(California)

 Messidor 2 (1748 images) 93% 87%
Our study

(India)

 1540 images 98.5% 93.5%
Referable DR
Gulshan et al8

(Google Inc, USA)

 EyePACS-1 (9963)

 

 

 90.3%

97.5%

 98.1%

93.4%
Messidor 2 (1748)

 

 87%

96.1%

 98.5%

93.9%
Wong et al9

(Singapore)

 Multiple sets of images 90.5% 91.6%
Ramachandran et al10 (New Zealand)

 

 ODEMS (382)

MESSIDOR (1200)

 84.6%

96%

 79.7%

90%
Our Study

(India)

 1540 images 99% 98.3%

Table summarising other studies

Leave a Comment