Skip to main content
Aurélien Vannieuwenhuyze

Doctoral research · CNAM · CEDRIC / ISID team

Clinical Trajectories

I develop methods to represent medical events over time, connect data with explicit knowledge, and keep AI outputs open to scrutiny.

01 / The research question

A clinical trajectory is more than a snapshot

In critical care, an isolated value only becomes meaningful in relation to what came before: a biomarker trend, the onset of a symptom, an administered treatment, an observed response, or a change in condition.

The challenge is to represent that history in a form that AI systems can use without losing temporal relationships or the ability to inspect the reasoning behind an output.

Research directions

Four complementary dimensions

01

Temporal representation

Structure clinical events, durations, sequences, and relations so that patient trajectories can be queried and analysed.

02

Explicit knowledge

Connect data with medical concepts, ontologies, and rules to preserve meaning beyond statistical correlations alone.

03

Neuro-symbolic AI

Investigate combinations of machine learning and symbolic reasoning to build models that are more interpretable and controllable.

04

Clinical usefulness

Evaluate not only predictive performance, but also the clarity, limitations, and decision-support value of the resulting information.

Research process

From a clinical question to a verifiable result

  1. 01

    Frame the question

    Define the events, temporal relations, and required information from a precise clinical question.

  2. 02

    Build the representation

    Transform longitudinal data into an explicit, queryable temporal structure connected to domain concepts.

  3. 03

    Learn and reason

    Study how learning models, knowledge graphs, and rules can cooperate without hiding their assumptions.

  4. 04

    Evaluate and explain

    Assess performance, robustness, reproducibility, and the understandability of results in the intended context of use.

Doctoral publication

A Conceptual Model for Discovering Implicit Temporal Knowledge in Clinical Data

Aurélien Vannieuwenhuyze · Nada Mimouni · Cédric Du Mouza

ER 2025 · volume published in 2026 · Lecture Notes in Computer Science, vol. 16190 · Springer · pp. 81–100

This first study proposes a conceptual model that transforms intensive-care data into an OWL-Time-compliant temporal knowledge graph. Using MIMIC-IV data and a bloodstream-infection use case, it shows how temporal relations that are not explicit in the source data can be made reproducible and queryable.

Temporal knowledge graph OWL-Time MIMIC-IV Intensive care

Open questions

The doctoral research programme

  1. 01

    How can irregular clinical events and their temporal context be represented faithfully?

  2. 02

    How can medical knowledge be integrated without freezing the model or hiding uncertainty?

  3. 03

    How can a prediction be related back to the patient trajectory and the factors that shaped it?

  4. 04

    How should the usefulness of an explanation be evaluated for healthcare professionals beyond technical plausibility?

Rigor and limitations

Research designed to remain verifiable

The work presented here concerns research and prototyping. It is not a validated medical device, a diagnostic tool, or a substitute for clinical judgment.

  • Assumptions, transformations, and evaluation criteria made explicit.
  • Reproducibility prioritised whenever data access and licences allow it.
  • No directly identifiable health data published on this website.
  • Clinical and external validation required before any operational use.

APPLIED RESEARCH PROJECT · 001

TENSYM.

Agentic AI & Temporal Neuro-Symbolic Reasoning for Clinical Trials

What did an AI agent know when it made its decision?

An experimental project on error propagation between AI agents, temporal rules and decision traceability in a fully simulated clinical trial.

Explore TENSYM

Applied research

Transferring research methods to concrete R&D problems

This research informs my applied work on longitudinal-data framing, knowledge graphs, temporal prototypes, and the explainability of healthcare AI systems.