Inspiration
Anemia is a silent epidemic among adolescents, especially girls, with millions going undiagnosed due to low awareness, lack of regular testing, and poor adherence to treatment. Traditional testing requires blood draws, which are invasive and discourage regular screening.
We wanted to build something that is non-invasive, accessible, and scalable — something that could fit right into the adolescent lifestyle without fear or stigma.
What it does
AnemiC is a smartphone-based tool that uses computer vision and deep learning to detect the risk of anemia from a simple eye photograph.
- 📸 Eye Scan Technology: Upload an eye photo, get an instant result.
- 🧠 AI-powered ensemble model: EfficientNetV2B0 + ResNet50V2 trained on medical datasets.
- 📊 Performance: 87% confidence score with 81.57% AUC.
- 🌍 Accessible: No needles, no labs, just a smartphone camera.
- 🎓 Awareness & Behavior Change: An integrated chatbot educates adolescents about anemia, supplements, and lifestyle changes in a friendly, engaging way.
How we built it
- Collected and pre-processed eye images from open-source medical datasets.
- Designed a CNN ensemble architecture (EfficientNetV2B0 + ResNet50V2) for feature extraction and classification.
- Trained and validated the model, optimizing for AUC and confidence score.
- Built a React + Tailwind frontend with an API integration for image upload and real-time predictions.
- Integrated a chatbot module to provide personalized education and reminders.
Challenges we ran into
- ⚙️ Training CNNs on limited medical datasets without overfitting.
- 📱 Designing a clean, professional UI that feels engaging to adolescents yet trustworthy for healthcare.
- 🔄 Ensuring fast inference so that the results are available instantly on smartphones.
- 🌍 Balancing technical accuracy with user-friendly explanations (adolescents may not understand medical jargon).
Accomplishments that we're proud of
- Achieved 81.57% AUC with a hybrid CNN approach.
- Built a non-invasive solution that can work with just a smartphone.
- Created a prototype with real-world usability — instant results, awareness, and education in one place.
- Showed how AI can be leveraged for public health impact in adolescent communities.
What we learned
- How to combine deep learning and healthcare responsibly.
- The importance of human-centered design — it’s not just about detecting anemia, but also about making adolescents care.
- Deploying AI models into scalable, tech-enabled solutions requires both engineering and design thinking.
What's next for AnemiC
- Expand dataset size with collaborations for higher accuracy.
- Deploy as a progressive web app (PWA) so it works even offline.
- Build regional language support for inclusivity across India.
- Partner with schools and NGOs to bring AnemiC to adolescent communities at scale.
- Explore integration with wearables for continuous, passive screening.
Technical Math Insight
To evaluate our model, we relied on the Area Under the ROC Curve (AUC) metric.
The formula is:
[ AUC = \int_0^1 TPR(FPR) \, d(FPR) ]
where
- ( TPR = \frac{TP}{TP + FN} ) (True Positive Rate)
- ( FPR = \frac{FP}{FP + TN} ) (False Positive Rate)
Our best model achieved:
[
AUC = 0.8157 \ (\approx 81.57\%)
]
Log in or sign up for Devpost to join the conversation.