CAIMed presents

CAIMed Meet-up: Trustworthy AI, Causality and Deep Learning in Medicine

14:00 - 17:30 29-11-2024 @ L3S Research Center

CAIMed Meet-up: Trustworthy AI, Causality and Deep Learning in Medicine

We are pleased to invite you to the upcoming CAIMed Meet-up on November 29, 2024, from 2:00 p.m. to 5:30 p.m. PM at the L3S Research Center in Hannover. The meet-up will focus on "Trustworthy AI, Causality and Deep Learning for Medicine", exploring the intersection of AI technologies and causal models in healthcare.

In recent years, AI has achieved remarkable advances, but creating trustworthy AI-algorithms that are fair, explainable, and robust remains a challenge. This meet-up will delve into how causal representation learning is helping to address these issues and shaping the future of AI-driven healthcare.

This event will feature a series of presentations from researchers within the CAIMed consortium, covering topics like AI in genomics, cancer prediction, and digital pathology.

We hope you can join us for an engaging discussion and networking opportunities.

Download Agenda


Agenda

2:00 p.m. – 2:10 p.m. Welcome & Introduction (Prof. Dr. techn. Wolfgang Nejdl, CAIMed)


2:10 p.m. – 3:30 p.m. Session 1: Causal Representation Learning and Foundation Models (CAIMed, Junior Research Group on AI & Causality)

1. AI Foundation Models in Genomics, Dr. Michelle Tang
  • We focus on single-cell RNA foundation models. Our initial work involves benchmarking recently published scRNA foundation models, specifically evaluating their performance in cell annotation and batch integration tasks. Additionally, we test our hypothesis that foundation models outperform baseline models on out-of-distribution data by generating our own datasets.
2. Exploring Causal AI Avenues in Medicine, Azlaan Mustafa Samad, M.Sc.
  • Literature review to find new avenues of Causal AI application such as Drug Discovery. Exploring currently available benchmark dataset for Causal Inference such as CausalBench, UK Biobank etc.
3. Cancer Type Prediction and Causal Hypergraphs, Johanna Schrader, M.Sc.
  • Cancer Type Prediction: We utilise image-based deep learning techniques to detect 13 distinct cancer types, along with a non-cancer class, from serum microRNA expression data. By fine-tuning our model, we achieve competitive results with limited resources.

  • Causal Hypergraphs: We conceptualize social interactions as hyper graphs, capturing complex group dynamics beyond simple pairwise relationships. By employing disentangled representations, we effectively model treatment assignment probabilities similar to propensity scores, enabling individual outcome predictions.

4. Large Language Models for Psychotherapy, Julian Laue, M.Sc.
  • Evaluating LLMs' knowledge of behavioural activation therapy compared to psychology students, using a questionnaire before and after providing educational material. LLMs outperformed humans on average in both pre- and post-tests, while both groups showed improvement. The study explores LLMs' potential in addressing Germany's therapist shortage and mental health support needs.

3:30 p.m. – 4:00 p.m. Session 2: RESIST: Remapping EIT signals using an implicit spatially-aware transformer (Dominik Becker, University of Göttingen)

  • The aim of electrical impedance tomography (EIT) is to reconstruct a body's internal conductivity distribution. To solve this inverse and ill-defined problem, we propose RESIST, which combines the location bias from implicit neural networks with the feature extraction power of transformer models.

4:00 p.m. – 4:30 p.m. Session 3: AI-based Digital Pathology in Oncology and Cancer Screening (Felipe Miranda Ruiz, M.Sc. and Prof. Dr. Niels Grabe, CAIMed / University Medical Center (UMG) Göttingen)

  • Using AI-based applications, we process Whole Slide Images for cancer screening and prognosis. We evaluate AI model performance and predictive robustness to image variability in multi-institutional patient cohorts (e.g. variability in staining/scanning). We propose strategies to address and improve robustness with the goal of implementing AI solutions in clinical practice.

4:30 p.m. – 4:35 p.m. Wrap-up / What’s coming next? (Prof. Dr. Marius Lindauer, CAIMed, University of Hannover)


4:35 p.m. – 5:30 p.m. Reception & Networking

Speakers

CAIMed is funded by the Ministry of Science and Culture of Lower Saxony with funds from the program zukunft.niedersachsen of the VolkswagenStiftung

Ticket(s)

This event has ended.
CAIMed

CAIMed is the Lower Saxony research center for artificial intelligence and causal methods in medicine. We develop innovative methods for improved, personalized healthcare and contribute to the management of widespread diseases such as cancer, cardiovascular diseases and infections. The combination of excellent locations in Lower Saxony for methodical AI research, data-intensive medicine, medical informatics and basic medical research creates a unique flagship project for research into AI and personalized medicine.

CAIMed relies on the linking of research data, clinical data and patient care data as well as the use of artificial intelligence and causal methods. This enables prevention, diagnostics, therapy and monitoring of therapeutic success to become more effective and efficient and the individual needs of each person to be better identified and served.

Contact CAIMed