Discover, Connect & Collaborate at TECHINNOVATION 2021

Automated Diagnosis of Retinal Image (normal/abnormal) using Deep Neural Network

Technology Overview

In ophthalmology, the three main eye diseases that can lead to blindness are age-related macular degeneration (AMD), diabetic retinopathy (DR) and glaucoma. Machine learning and deep learning are two emerging techniques that can be used for screening and diagnosis of retinal diseases using fundus images. Machine learning algorithm is not optimal for automatic diagnoses, as the user needs to define each feature for the algorithm to learn and diagnose the diseases.

This innovation offers an automated diagnostic solution for retinal health based on fundus image and deep learning technology. The solution contains an improved robust Convolutional Neural Network (CNN) model that provides higher accuracy and sensitivity using deep learning technique to automatically classify retinal fundus images of age-related macular degeneration (AMD), diabetic retinopathy (DR), glaucoma, into abnormal and normal classes. The network also can be run on any computing platform, delivering instant result for clinicians and patients. The system can be deployed at polyclinics and community centers for mass retinal health screening.

The technology provider is looking for industry partners to commercialize the technology in overseas eye hospitals and healthcare institutions. The target markets include Malaysia, Thailand and India.

Technology Features, Specifications and Advantages

The developed 10-layered neural network is able to automatically classify images of age-related macular degeneration (AMD), diabetic retinopathy (DR) and glaucoma as abnormal, and images of normal subjects as normal. The input image for the system is of size 180 x 270 pixels. The network uses different sized kernels to interpret the input fundus image, thereafter the feature maps are being concatenated for analysis.

The system was developed and tested on a total of 2986 images (collecting from various sources). ‘ADAM’ optimizer was used to train the net, and achieved an accuracy of 95.24% on a set of 1492 images. The sensitivity and specificity was 91.67% and 96.81% respectively. The software device has been developed in accordance with IEC62304. The developed network is commercially ready for deployment to any computing or mobile devices.

Potential Applications

The market for Deep Learning market by end-user for the healthcare industry is expected to grow at the highest CAGR of 55.30% during the forecast period, from USD $0.29 billion in 2017 to USD $3.93 billion by 2023. Significant developments in AI, together with the growing penetration of various AI technologies and products in countries such as Singapore, Indonesia, the Philippines, Taiwan, Malaysia, Thailand, Australia, and New Zealand, contribute to growth in the rest of Asia-Pacific markets.

Key customer segments include:

  • Medical institutions and eye hospitals: These healthcare service providers own the infrastructure, I.T. system and the provision of medical specialists to provide quality and affordable medical treatment to the public. The automated diagnosis solution can be deployed at any clinical facility for the mass screening and routine screening of the fundus for referral to medical eye specialists
  • Telemedicine, mHealth Service providers: These virtual service providers specialize in providing expert opinions to patients. These providers can adopt our solution to give the patients a comprehensive picture on their conditions and offer guidance on the necessary steps that for disease management and recovery
  • Pharmacy retail outlets and community centres – These are channels where the general public have 24/7 access to the automated diagnosis solution as self-service preliminary assessment of their eye conditions. The assessment can thereby serve as the basis for referral to eye specialists for follow up treatment, where appointments with eye specialists can be done online.

Customer Benefit

  • Fast and reliable diagnosis
  • Reduces clinician’s workload and improves productivity of healthcare practitioners.
  • Network is compact (small)
  • Readily to be deployed on any computing or mobile devices.
Contact Person

Dickson Phuan


Ngee Ann Polytechnic

Technology Category

  • Healthcare
  • Diagnostics, Telehealth, Medical Software & Imaging

Technology Readiness Level


Deep Learning, Age-Related Macular Degeneration, Diabetic Retinopathy, Glaucoma, machine learning, fundus, neutral network, Automated diagnosis