Diabetic retinopathy (DR) is a disease with an increasing prevalence and the main cause of blindness among working-age population. The risk of severe vision loss can be significantly reduced by timely diagnosis and treatment. The current process of testing for diabetic retinopathy is laborious and often inefficient and the scope of detecting the diseases in the early stages are completely unexploited. There exists a need for a powerful Automated retinal image analysis for early stage Diabetic retinopathy detection

What it does

The tool analyzes the retinal image of a patient to detect whether the patient is affected with Diabetic Retinopathy and if yes, how severe (on a medical scale of 1 to 4; 4 being highest). It is the ultimate tool to aid in precisely detecting the early stages of diabetic retinopathy. .

How I built it

The entire application was developed using the Predix infrastructure . The application was build in different layers UI using JavaScript angular js, html , Postgre sql database, Data science algorithm in python. We leveraged powerful deep learning algorithms inspired from the way the humans learn. Convolutional Neural Networks were leveraged to learn from retinal images of several hundred of different severity levels .

Challenges I ran into

The Training process for the deep learning algorithms were extremely resource intensive, our codes kept running for over 15 hours with no intermediate outputs from predix analytics runtime . With limited visibility into the analytics processing and time we ran in to challenges of optimizing the resource intense algorithms. Our attempt to improve computational efficiency by increasing instances and RAM was also not successful .

Accomplishments that I'm proud of

Making end to end predix Micro application In the short time using various predix components such as Postgreql,Predix seed application(UI),Predix UAA,Predix analytics catalog, Predix Analytics Runtime . deploying in to the Predix platform. Building the deep learning algorithm for the computer vision in a day . Falling early and finding out alternatives quickly to achieve the goal .

What I learned

How to plan end execute the things in short cycles . How to cut the corners and still make wonderful applications in 24 hours . How to work collaboratively to get things done as team . Constant motivation and energy to achieve the goal .

What's next for Trojan - Automated Diabetic Retinopathy Detection

Enhancing the algorithm as a real-time service using Analytics micro service to be monetized in the predix platform . Improving the algorithm into self-learning algorithms that can be completely automated and improved incrementally without human interventionExtending the scope to other industries like Manufacturing to learn to predict and detect faults. Eg - detecting manufacturing defects using real-time image analysis

Share this project: