This PhD project is a collaboration between AZ Delta/RADar Learning & Innovation Center, Sciensano, Intermutualistisch Agentschap (IMA) and UHasselt. It investigates how artificial intelligence can learn meaningful patient representations from longitudinal health data, enabling earlier detection of rare diseases.

---

Objective

Patients with rare diseases often experience a years-long diagnostic odyssey before receiving an accurate diagnosis. The project explores how deep learning can capture diagnostic signals hidden in patient trajectories, supporting earlier identification of at-risk patients.

Methodology

The project applies patient representation learning on longitudinal electronic health records and administrative data. Particular attention is given to addressing the strong data imbalance between rare disease and non-rare disease patients.

Impact and future directions

This research could support the early flagging of at-risk patients and help optimize referral for diagnostic evaluation. Earlier detection may reduce the psychological burden associated with diagnostic uncertainty and enable faster access to appropriate care.

General info & contact

Keywords (#): rare diseases, artificial intelligence (AI), patient representation learning, electronic health record (EHR) embeddings, imbalanced data

RADar project research lead: M.Sc. Lize Devolder

Academic Promotor: Prof. Dr. Ward Schrooten (UHasselt)

Industrial Promotor: Prof. Dr. Peter De Jaeger (AZ Delta/RADar Learning & Innovation Center), Prof. Dr. Sofie De Broe (Sciensano), Dr. Sofie Vanassche (Intermutualistisch Agentschap, IMA)

Timeline: 2026-2030

Status: Initial investigation of embedding techniques