About the Project
💡 What Inspired Us
Anaemia remains a major health challenge, especially for women, adolescent girls, and children in rural and low-resource communities. We were inspired by a simple question: Can early anaemia-risk screening be made as accessible as taking a photo?
Traditional screening often requires a blood test and access to a healthcare facility. We wanted to explore a non-invasive, affordable, and offline-first approach that could help health workers identify people who may need further testing.
🛠️ How We Built It
Our solution, AI-Based Ocular Anaemia Screening, uses a smartphone camera to capture an image of the inner lower eyelid. A printed colour card is included in the image to help standardise colour and reduce variations caused by different lighting conditions and phone cameras.
The system follows a simple pipeline:
Capture → Detect → Calibrate → Predict → Advise → Follow-up
We developed the AI component using Python, TensorFlow/Keras, transfer learning with MobileNetV2, and OpenCV for image cropping and colour correction. The mobile application is designed with Flutter, with TensorFlow Lite enabling on-device inference and SQLite for local data storage.
The app is designed to work offline, providing a Low/Medium/High risk result in approximately one minute. Bengali, Hindi, and English guidance makes the workflow more accessible, while referral and re-screening reminders support follow-up care.
📚 What We Learned
Building this project taught us that developing an AI healthcare solution is about much more than training a model. Data quality, lighting conditions, skin-tone diversity, validation, privacy, and usability are equally important.
We also learned how transfer learning and lightweight models can make AI more practical for resource-constrained devices, and how an offline-first design can be valuable where reliable internet connectivity is not guaranteed.
⚠️ Challenges We Faced
One of our biggest challenges was the variability of real-world images. Lighting, blur, eyelid position, skin tone, age, and other eye conditions can affect the captured image and potentially influence the prediction.
Another challenge is the limited availability of Indian field images paired with measured haemoglobin (Hb) values. To address these limitations, our proposed approach includes image-quality checks, colour calibration, data augmentation, diverse validation data, and collaboration with healthcare workers for field validation.
Most importantly, we designed the system as a screening tool, not a diagnostic tool. Users identified as being at risk should be referred for appropriate Hb testing.
🌍 Our Goal
Our goal is to make preliminary anaemia-risk screening more accessible, affordable, and practical, particularly for communities where laboratory-based screening may be difficult to access.
One photo can start the journey from screening to timely care.
Log in or sign up for Devpost to join the conversation.