Publication in npj Digital Medicine

Personalized Federated Learning for MS Disability Progression Prediction

• Published: July 24, 2025 in npj Digital Medicine
• Title: “Personalized federated learning for predicting disability progression in multiple sclerosis using real-world routine clinical data”
• Link: Read the paper

Why This Paper is Special

This publication represents one of the projects I am proudest of. From the first brainstorming sessions with co-authors, to writing and submitting the proposal to The MSBase Foundation to access real-world data, securing ethical approvals, running experiments with Flower Labs on the VSC Vlaams Supercomputer Centrum, and bringing it all the way to publication, every step was executed with care, passion, and patience.

What makes it special: This is the first federated learning study in multiple sclerosis using routine clinical data. That makes the journey even more meaningful.

Key Findings

• Dataset: 26,000+ patients, 283,000+ prediction episodes, 146 centers in 32 countries
• Task: Predicted 2-year disability progression in multiple sclerosis
• Results: Personalized FedProx and FedAVG achieved ROC-AUC scores of 0.8398 ± 0.0019 and 0.8384 ± 0.0014, respectively
• Impact: Established personalization as important for scalable, privacy-aware clinical prediction models

Acknowledgments

I learned a lot from working with Edward De Brouwer and valued his active contributions and steering throughout. I feel lucky to have had Yves Moreau and Liesbet M. Peeters as supervisors, whose guidance and mentorship shaped the project in important ways.

Thank you to MSBase for backing the idea, VSC for providing computing resources, and Vlaams AI Onderzoeksprogramma / Flanders AI Research Program for their support in making this project possible.

This one will always stay with me.