Inspiration

Maternal anemia remains an important healthcare challenge, particularly in rural and underserved communities where access to laboratory-based screening can be limited. We were inspired by the possibility of using smartphones and AI to make preliminary anemia-risk screening more accessible without requiring a blood draw at the point of screening.

Instead of focusing only on detection, we wanted to create a complete support workflow — screen → assess → provide nutrition guidance → refer → follow up.

What it does

HEMOSIGHT AI is a proposed AI-based maternal anemia screening and nutrition follow-up platform.

The system combines:

A conjunctival eye image
A finger PPG signal captured using a phone camera and flash
Symptoms and pregnancy information, including age, gestational weeks and trimester

These inputs are processed into 13 features and combined using a Random Forest model to classify anemia risk as LOW, MODERATE, or HIGH, along with confidence and class probabilities.

The result can then support:

Nutrition guidance → IFA tracking → ASHA referral → ANM review → Doctor review

The platform is designed around five roles: Admin, Doctor, ASHA, ANM and Mother.

How we built it

For the proposed implementation, we designed a multimodal AI pipeline using PyTorch and EfficientNet-B0 for conjunctival image analysis, Python-based DSP with Butterworth filtering and peak detection for PPG processing, and scikit-learn Random Forest for multimodal fusion.

The application architecture uses React, Spring Boot microservices, API Gateway and Eureka, with MongoDB and Oracle XE for data management. Security is planned through JWT, BCrypt, OTP and role-based access control.

The core pipeline is:

Eye image + PPG + symptoms/pregnancy data → 13-feature fusion → Random Forest → Risk classification → Nutrition & Referral

Challenges we ran into

The major challenges were designing a solution that could work with real-world smartphone conditions and still remain safe for healthcare use.

Some key challenges include:

Poor lighting, blur and glare in eye images
Noise and movement during PPG capture
Combining image, signal and clinical data
Protecting sensitive patient information
Making the workflow simple for frontline healthcare workers
Ensuring that an AI screening result is not treated as a medical diagnosis

To address these, we proposed image-quality checks, PPG filtering and signal-quality metrics, secure authentication, and escalation of moderate/high-risk cases to healthcare professionals.

Accomplishments that we're proud of

We are proud of designing HEMOSIGHT as more than a prediction model. The idea connects AI-based screening with an actual healthcare workflow.

Our key accomplishments include:

Designing a 13-feature multimodal AI approach
Combining eye image, PPG, symptoms and pregnancy information
Designing a five-role healthcare platform
Connecting screening results with nutrition and follow-up
Designing an ASHA → ANM → Doctor referral pathway
Building the concept around smartphone-based, non-invasive screening
Establishing a clear validation roadmap with healthcare professionals and laboratory Hb measurements

What we learned

We learned that healthcare AI is not only about achieving a prediction. A meaningful healthcare solution must also consider data quality, privacy, clinical safety, usability and the complete care workflow.

We also learned the importance of combining multiple signals instead of depending on a single indicator. HEMOSIGHT brings together visual, physiological and clinical information to create a more comprehensive anemia-risk screening approach.

What's next for HemoSight

Our next step is to move from an idea and prototype concept toward clinical validation.

The planned roadmap is:

Validate the AI screening against laboratory Hb measurements at a partner PHC.
Conduct a field pilot with ASHA and ANM teams.
Improve the models using validated real-world data.
Test usability and reliability under different smartphone and environmental conditions.
Gradually scale the solution across districts.

Our long-term vision is to make preliminary maternal anemia-risk screening more accessible while keeping healthcare professionals at the center of clinical decision-making.