London 2025

As part of the Network’s key performance indicators (KPIs), a meeting was held between the two coordinators, Janna Hastings and Ernesto Jimenez-Ruiz. This visit also provided an opportunity to go to St George’s Hospital, where they met with Franklyn Howe and presented the Network’s work to a group of clinicians.

Agenda

Day 1 (afternoon)

• Meeting Janna and Ernesto: City St George’s, University of London (Tooting campus)

Day 2

10:45 – 12:00 — Meeting

Participants:

• Janna, Ernesto, Franklyn, Ian and Tom
  Focus: Potential use of neurosymbolic AI to enhance the automated analysis of multimodal MRIs.

12:00 – 13:00 — Session with Clinicians

• Presentations by Janna and Ernesto
• Q&A discussion

13:00 + — Lunch & Networking

Janna’s presentation

Title: Catching the Wave: Neuro-symbolic Approaches Help Manage Medical Evidence

Abstract:

The presentation explores how neurosymbolic AI—the combination of machine learning and ontological reasoning—can help manage and synthesise the rapidly expanding medical evidence base. With over a million biomedical papers published yearly, traditional systematic reviews are increasingly unmanageable, and while large language models (LLMs) show promise for automating evidence screening and data extraction, their reliability and interpretability remain limited.

The talk demonstrates practical evaluations of open-source LLMs for literature screening and structured data extraction, showing variable performance across models and topics. It then highlights how ontologies provide structured, consensus-based definitions that can improve transparency, reproducibility, and reasoning in medical AI systems.

Ernesto’s presentation

Title: Knowledge Graphs in the era of Large Language Models

Abstract:

Large language models (e.g., ChatGPT) are leading to impressive results on text generation. There are however important issues with respect to bias, fairness, copyright violation, misinformation and explainability. The combination with symbolic knowledge (e.g. knowledge graphs) may mitigate these issues. At the same time large language models are becoming essential on the creation and extension of symbolic knowledge, facilitating their application at scale. The presentation provides a brief summary of the opportunities of combining (Large) Language Models with knowledge graphs.