Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Natural Language Processing (NLP) plays a crucial role in analyzing medical text for risk detection and improving patient-doctor communication. However, working with sensitive clinical data presents significant challenges, particularly in anonymization and patient privacy protection.

This talk will focus on privacy-preserving NLP, specifically anonymization techniques that go beyond direct identifiers to address risks from implicit information. I will discuss methods and challenges of de-identifying medical text while preserving its utility for downstream tasks such as risk prediction and clinical decision support. Beyond privacy, I will address how NLP can be leveraged for early risk detection: From multilingual adverse drug reaction detection to mental health risk assessment on social media and relapse prediction in psychotherapeutic settings.

Referent: Dr. Lisa Raithel

Kurzbiographie

Dr. Ratihel is a post-doc at Technische Universität Berlin at the Quality and Usability Lab and BIFOLD and a guest researcher at DFKI GmbH. She obtained her master’s degree at Universität Potsdam in Computational Linguistics (B.Sc. in Computational Linguistics, M.Sc. in Cognitive Systems). Then worked as a software engineer before transitioning back to academia for a double degree PhD program (cotutelle) at TU Berlin and Université Paris-Saclay. She was supervised by Prof. Sebastian Möller and Pierre Zweigenbaum, Directeur de Recherche CNRS. Her doctoral research focused on cross-lingual information extraction for the detection of adverse drug reactions. During that time, she spent one year at LISN in Orsay, France (2021 - 2022) and three months at the Social Computing Lab at NAIST in Nara, Japan (2023). In February 2024, she successfully defended her thesis at TU Berlin.

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

The body of published biomedical literature is growing at a rate that challenges manual management. Large Language Models (LLMs) enable large-scale processing of textual information and have the potential to enable a step change in how we can use evidence and research through detailed and flexible automation of information extraction. In this presentation, I will present recent progress towards harnessing LLMs to accelerate systematic reviewing and the translation of evidence into personalised treatments, with applications in mental health and oncology.

Referent: Prof. Dr. Janna Hastings

Kurzbiographie

Janna Hastings was born in Cape Town, South Africa where she completed undergraduate studies in Mathematics and Computer Science. Thereafter, she moved to Cambridge, UK to join the Cheminformatics and Metabolism group at the European Bioinformatics Institute (2006-2015) and obtained her PhD in Computational Biology from the University of Cambridge (2015-2019) studying the role of metabolism in healthy aging using multi-omics data and a time-series modelling approach. Since August 2022 she is Assistant Professor of Medical Knowledge and Decision Support at the Institute for Implementation Science in Health Care, Faculty of Medicine, University of Zurich, and Vice-Director of the School of Medicine at the University of St. Gallen. She is also an Associate at the Centre for Behaviour Change at University College London, and Group Leader of the Swiss Institute for Bioinformatics. The focus of her current research is on AI in medicine.

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Human trabecular bone is a structural, calcified tissue of struts and plates with cavities containing bone marrow. Determining its mechanical properties can support the design of patient-specific implants, e.g., for total hip- or knee arthroplasties. The effective stiffness is obtainable based on the direct discretization of microfocus computed tomography (mCT) data. However, the Finite Element Analysis (FEA) based analytical approach is computationally expensive and requires high-performance computing (HPC) resources that are unsuitable in daily clinical context. Our research introduces an artificial intelligence (AI) centric method to determine the effective stiffness of human trabeculae with sufficient accuracy at a fraction of the computational cost of current analytical implementations. This talk furthermore shows the data, training process, and validation of the biomechanical results with the analytical state-of-the-art method as the ground truth. Furthermore, we will describe the limits and outlook for contributing to patient-specific mechanical characterizations.

Referent: Johannes Gebert

Kurzbiographie

Johannes Gebert has about 10 years of experience as a design, structural, and software engineer. His graduate studies connected him to the High-Performance Computing Center Stuttgart (HLRS). He joined as a PhD student of Prof. Resch to work on the interface of domain-specific research with HPC. The PhD report, submitted in June 2024, describes a novel method to calculate anisotropic elasticities of human trabeculae in vivo. A research visit at the Innovative Computing Laboratory (ICL) at the University of Tennessee, Knoxville (UTK), strengthened his background in HPC. He joined HLRS in a permanent position to focus on innovative computing machinery to accelerate the computational capabilities of future research.

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referentin:

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Surgical Data Science is a rapidly evolving interdisciplinary field that aims to improve the quality, safety, and value of interventional healthcare through data acquisition, organization, analysis, and modeling. Over the past decade, advances in deep learning, computational power, and collaborative medical data initiatives have driven significant progress in computer vision and natural language processing within the surgical domain.

This talk will explore the emerging role of foundation models, large models pre-trained on broad data using self-supervised learning, and their ability on a wide range of clinically relevant tasks, often with minimal labeled data. I will review recent surgical image and video-based foundation model developments, discuss key downstream applications, and highlight common pitfalls in benchmarking their performance. Additionally, I will present current strategies for leveraging large language models (LLMs) to standardize surgical metadata and improve data interoperability at scale.

Referent:  Dr. Ömer Sümer

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referent:

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referent:  Dr. Fabian Horst

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referent:

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referent:

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referent:

Kurzbiographie

Veranstaltungsort: Lecture hall N2045 (Faculty of Applied Computer Science)

Abstract

Referent:

Kurzbiographie