MedOracle – Decentralized Medical Report Verifier
MedOracle: Medical Reports, Verified on Blockchain
About
The problem it solves 🧩 The Problem it Solves Modern healthcare generates a vast number of medical test reports daily—many of which are misunderstood by patients, misused in falsified documents, or improperly handled in insecure systems. 🚫 The current pain points: Non-technical patients often don’t understand what their test results indicate. Manual verification of reports (e.g., for job screenings or insurance) is time-consuming and prone to human error. Tampering and forgery of digital reports is a serious issue in institutions with limited checks. Lack of data privacy—medical records are stored in centralized systems vulnerable to leaks or misuse. 💡 What MedOracle Offers MedOracle is an AI-powered + blockchain-backed solution that helps users: ✅ Instantly Understand Test Results Upload a lab report (image or PDF) Extracts key medical features automatically (e.g., BP, sugar levels) Uses an ML model to predict health risk level in plain terms 🔐 Securely Store Medical Proofs on Web3 Converts test data into a tamper-proof record Stores a hash on IPFS, with an immutable reference recorded on-chain via a smart contract This makes medical records verifiable, traceable, and privacy-preserving ⚙️ Use Cases: Patients seeking clear explanations of their reports Hospitals needing fast, AI-assisted triaging Employers and insurance agencies verifying authenticity of health records Researchers needing anonymized, decentralized medical data
How it's made
Challenges we ran into ## 🐞 Challenges I Ran Into Building **MedOracle** involved tackling several technical and integration challenges: ### 🔗 1. Connecting Web3 with Python (Flask) Unlike JavaScript-based dApps, integrating blockchain functionality in a Python backend was tricky. Managing `web3.py`, contract ABIs, and triggering smart contract functions from Flask required careful coordination. **Solution:** Created a `contract_connector.py` module to handle smart contract loading and transaction logic separately, improving modularity and clarity. --- ### 🧠 2. ML Model Compatibility Issues An early issue involved a `ValueError` when loading a `DecisionTreeClassifier` trained with `scikit-learn 1.0` in an environment running `1.7`. This caused the app to break during model loading. **Solution:** Re-trained the model using the matching `scikit-learn` version and saved it in a fresh `pkl` file that matched the current environment. --- ### 📄 3. OCR from Uploaded PDFs Initially, the app only supported image input. But many users had test reports in PDF format. **Solution:** Used the `pdf2image` library to convert PDFs into images before OCR extraction, allowing the same processing pipeline for both input types. --- ### 🧪 4. IPFS Upload + Metadata Mapping Storing results on IPFS was easy, but linking the CID back to user reports and blockchain records needed care. **Solution:** Wrote custom JSON objects with extracted features + prediction results and uploaded them using `web3.storage`. The returned CID was then registered via the smart contract. --- ### 🧬 5. Time Crunch! Building the entire flow — frontend, backend, ML, OCR, blockchain, IPFS — within a tight timeframe was tough. **Solution:** Planned a modular structure early on, worked in small testable chunks, and reused proven libraries to accelerate development. ---