RetinaScan AI

LOGIN PAGE
HOME PAGE 1
HOME PAGE 2
DETAILS FILLING 1
DETAILS FILLING 2
IMAGE ADDING
ML MODEL PREDICTING
PREDICATION PERCENTAGE
RESULT PAGE
FINDING STAGES
FINAL REPORT
RESULT PDF

Inspiration

Diabetic Retinopathy (DR) can cause permanent vision loss when it is detected late. We were inspired by the difficulty faced by people in rural communities, where access to ophthalmologists and specialized eye-screening equipment can be limited. This led us to develop RetinaScan AI, an explainable AI-based solution for early diabetic retinopathy screening.

What it does

RetinaScan AI analyzes retinal/fundus images and detects signs of diabetic retinopathy. It classifies the condition into different severity levels:

Normal → Mild → Moderate → Severe → Proliferative

Unlike a system that only provides a prediction, RetinaScan AI also explains the result by highlighting the retinal regions that influenced the AI's decision. It provides a simple and understandable result and can recommend specialist referral when required.

How we built it

The proposed system follows this workflow:

Retinal Image → AI Model → DR Detection → Explainable Result → Severity & Referral

A retinal image can be captured using a low-cost fundus camera or a smartphone-assisted setup. The image is preprocessed and analyzed using a CNN/EfficientNet-based AI model. Grad-CAM is then used to highlight the retinal regions influencing the prediction.

The final output provides the detected condition, severity level, visual explanation, and referral recommendation.

Challenges we ran into

One of the major challenges was making the AI result understandable rather than providing only a disease prediction. We addressed this by incorporating Explainable AI using Grad-CAM.

Another challenge was designing the solution for rural environments, where expensive equipment, specialist availability, and reliable connectivity may be limited. This motivated our focus on a low-cost, accessible, and scalable screening approach.

Accomplishments that we're proud of

We developed a complete concept for an AI-powered diabetic retinopathy screening system that combines disease detection, severity classification, and explainability in a single workflow.

We are particularly proud of incorporating Grad-CAM so that the system can show the retinal regions influencing its prediction, making the result easier for healthcare workers and patients to understand.

What we learned

Through this project, we learned how AI and computer vision can be applied to medical image analysis. We explored CNN-based image classification, retinal image processing, severity classification, and Explainable AI using Grad-CAM.

We also learned that developing a healthcare AI solution requires more than accuracy. Accessibility, interpretability, and the ability to support healthcare workers are equally important.

What's next for RetinaScan AI

Our future plans include smartphone-based retinal screening, regional-language AI explanations, telemedicine and cloud integration, and offline/low-network support for remote areas.

We also aim to explore improved AI models for better accuracy and earlier-stage detection, with the long-term goal of making RetinaScan AI suitable for deployment in rural clinics, health camps, and mobile screening units.