HelixMatch - Smart Compatibility Engine

HelixMatch: Our AI engine intelligently delivers precision blood matches for Thalassemia patients facing rare types or alloantibodies, revolutionizing care and ensuring life-saving transfusions.

Comment

Inspiration

Our journey with HelixMatch was deeply inspired by the profound challenges faced by Thalassemia patients in India. While standard blood donation systems are vital, they often fall short for individuals who, due to repeated transfusions, develop alloantibodies. This makes finding a compatible donor a needle-in-a-haystack problem, leading to dangerous delays and immense stress for families. We realized that simply matching by ABO/Rh isn't enough; a deeper, more precise compatibility check at the genetic and cellular level is desperately needed. We were driven by the vision of leveraging cutting-edge AI to bridge this critical gap, ensuring that no patient is left without a life-saving match.

What it does

HelixMatch - Smart Compatibility Engine is an AI-powered solution designed to revolutionize blood donation and transfusion management, especially for Thalassemia patients with complex matching needs. Our system: Provides Precision Matching: It intelligently analyzes detailed donor and patient blood antigen profiles (like Duffy, Kidd, Kell, MNS, etc.) to find the most compatible matches, even for rare blood types or those with alloantibodies. Simulates Real-World Integration: The prototype demonstrates how it would seamlessly integrate with existing national blood management systems. We've simulated fetching comprehensive donor data from Blood Warriors' Blood Bridge database and processing patient requests originating from e-RaktKosh, showcasing a realistic operational flow. Offers AI-Powered Insights: For each matched donor, users can request an instant, AI-generated insight into the significance of that donor's unique blood antigen profile, providing crucial information for medical professionals. In essence, HelixMatch acts as a vital bridge, connecting patients to their ideal donors with unprecedented accuracy and speed.

How we built it

We built HelixMatch as a robust, two-part prototype: a core AI backend and an interactive web frontend. The AI Core (Python/PyTorch): At the heart of HelixMatch is a Siamese Neural Network, built using PyTorch. This network is specifically designed for similarity learning. We trained this model on a synthetic dataset of donor and patient antigen profiles. The model learns to encode these complex profiles into compact numerical representations (embeddings). To teach the model what constitutes a "match" versus a "non-match," we employed a Contrastive Loss function. This function penalizes the model when similar profiles are far apart in the embedding space and when dissimilar profiles are too close. The Python script demonstrates how, given a patient's profile, the model can efficiently compute compatibility scores with all available donors and rank them from best to worst match.

The Web Frontend (React/Tailwind CSS): We developed a dynamic, multi-page web application using React.js and Tailwind CSS for a responsive user interface. The frontend integrates with national blood management systems by fetching donor data from Blood Warriors' Blood Bridge database and processing patient requests from e-RaktKosh, complete with realistic loading delays. A key feature is the integration with the Gemini API: when a user clicks, the frontend sends a donor's profile to the LLM to generate concise, medically relevant AI insights, showcasing immediate, actionable information.

Challenges we ran into

One significant challenge was the inherent data sparsity and imbalance when dealing with rare blood antigen profiles. Training a Siamese network effectively requires diverse and representative pairs, which are difficult to acquire for extremely rare combinations. We addressed this by carefully designing our synthetic data generation process to capture the complexity and rarity, and by focusing the model on learning robust similarity metrics from limited examples. Another critical hurdle involved translating complex clinical compatibility criteria into a quantifiable metric for our AI model. Defining what "similar" truly means for a Siamese network, especially for alloimmunized patients, required careful consideration of medical guidelines and iterative refinement of our compatibility scoring logic.

Accomplishments that we're proud of

I am incredibly proud of several accomplishments with HelixMatch: Functional AI Core for Precision Matching: Successfully developing and demonstrating a working Siamese Neural Network prototype capable of learning and identifying complex blood antigen compatibility, a crucial step towards precision medicine. Seamless System Integration: Effectively showcasing the integration with critical existing systems like Blood Warriors and e-RaktKosh, proving the practical applicability and ease of adoption for HelixMatch. Innovative AI Insight Feature: Integrating the Gemini API to provide on-demand, intelligent insights into donor profiles, adding a unique and valuable dimension to the matching process for medical professionals. Impactful Prototype Development: Delivering a comprehensive prototype that clearly conveys the solution's potential to address a vital social challenge.

What we learned

This hackathon provided invaluable learning experiences: Applied ML for Medical Data: Deepened our understanding of Siamese Neural Networks and contrastive loss, particularly when working with sparse and imbalanced medical datasets. Bridging Clinical & AI Domains: Gained practical experience in translating complex clinical requirements into actionable machine learning problems. Effective Prototyping: Learned to build impactful prototypes that demonstrate core functionality and integration potential efficiently. Robust API Interaction: Acquired hands-on experience with reliable API integration, including error handling and retry mechanisms.

What's next for HelixMatch - Smart Compatibility Engine

HelixMatch has immense potential to evolve into a transformative tool for precision blood matching. Our immediate next steps focus on scaling our impact: Full-Scale Deployment & Live Integration: Our priority is to transition from this prototype to a robust, real-time system. This involves building a scalable backend to host our AI, establishing direct API integrations with Blood Warriors and e-RaktKosh for seamless data exchange, and deploying on a secure cloud platform like Microsoft Azure. Advanced AI & User Features: We aim to enhance our AI with predictive donor availability based on historical patterns. Concurrently, we'll develop features like gamified "Hero Donor" status to boost recurring participation and enable patients to securely save compatibility reports for future emergencies. Broader Impact & Expansion: Beyond Thalassemia, we envision applying HelixMatch's core AI matching capabilities to address compatibility challenges in other critical medical areas, extending its life-saving potential.

Built With

Share this project:

Updates