The problem MedOracle – Decentralized Medical Report Verifier 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

Challenges I 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.