This project explores a privacy-preserving decentralized federated learning system where hospitals collaboratively train local models while keeping patient data on-site, with a smart contract (Orchestrator) coordinating and validating the aggregation process.
See it to believe it! Explore how MedXChAln transforms healthcare AI through decentralized intelligence.
The following screens represent the web dashboard used to manage the federated learning lifecycle, monitor hospital nodes, and verify blockchain transactions.
| Overview & Analytics | Model Management |
|---|---|
 |
 |
| High-level system health and metrics. | Tracking FL rounds and aggregation. |
| Hospital Management | Diagnosis Support |
|---|---|
 |
 |
| Node monitoring and local logs. | Clinical AI-assisted diagnostics. |
| Security & Traceability |
|---|
 |
| Blockchain verification and IPFS CID history. |
- Local model training at multiple simulated hospitals
- Global model aggregation completed
- Multiple federated learning rounds executed
- Global model tested using
global_predict.py - Explainable AI (XAI) is added to make sure the training process is accurate (training the model is black box, in order to understand how the training is done XAI added via
global_explain_lime.py)
- Docker-based setup for hospitals and services
- Isolated environments for reproducibility
- IPFS nodes running inside Docker containers
- Model files can be added and retrieved manually
- IPFS integration is fully automated now
- Kubo-related behavior are understood.
- CID management and synchronization between nodes is done.
- Local Ethereum network running with Hardhat
- Smart contracts successfully deployed
- Model update logic prepared on-chain
- Contract is written and deployed to make sure autherized hospitals only participate
-
1- Decentralization
-
Full decentralization (removing central server/global model)
-
Monitoring network traffic and model size
-
Encrypting model updates before sharing
-
2-Hyper ledger Private Network
-
Moving from Hardhat open Network to Private permissioned network.
- Python, PyTorch
- Federated Learning (FedAvg)
- Explainable AI (XAI)
- IPFS
- Private Hyperledger Network
- Git
- Enhance the Keyway Orchestrator with advanced on-chain validation logic.
- Improve privacy through stronger encryption of model updates.
- Make the model more robust against adversarial attacks and other AI-based threats.
- Seham Hakim Othman
- Hatice Kübra Selvi