Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

Ever since the publication of ChatGPT, natural language processing (NLP) and large language models (LLM) are popular topics everybody has at least heard about.
Interestingly, there was not much development and interest in the German clinical community until this recent development (same as digitalisation in general). Now, everybody wants to be the first in utilising this (more or less) new technique. However, most of the datasets and models available are focused on English and the adaption to German or other languages and dialects is not a trivial task.
This talk will give an overview about the challenges and possible solutions in German clinical language processing, as well as the potentials and risks of recent developments in NLP research.

 

Referentin: Luise Modersohn, M.Sc. (Institut für KI und Informatik der Medizin, TU München)

 

Kurzbiographie

Luise Modersohn studied Bioinformatics in Jena. After completing her Master's degree she has worked as a research assistant at the Computer Vision Group at the Friedrich Schiller University of Jena for four years. Since 2018, she has been Research Assistant at the JULIE Lab in Jena. Since last year, she is leader of the DIFUTURE Junior Research Group De.xt at the Technical University of Munich. Her research interests are natural language processing with clinical context, hypergraphs and graph theory, bioinformatics and computational biology as well as computer vision and machine learning.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

The detrimental effects of mental stress on human health have been known for decades, and it has now developed into a major concern in our modern society, where it is regarded as a rising problem and an inevitable part of our everyday life. Psychological stress was shown to be separated into two types: acute and chronic. If stress is not recognized early, it can lead to a variety of diseases and major health problems, including hypertension and coronary disease, irritable bowel syndrome, gastroesophageal reflux disease, generalized anxiety disorder, and depression. In this talk, we will mention some experiments for recognizing stress by using unobtrusive wearables in different environments such as laboratory, real-life events and in the wild. Privacy preserving stress recognition approaches will be also explained. We will also talk about methods for alleviating stress and their performances for decreasing stress levels.

 

Referent: Dr. Yekta Said Can (Lehrstuhl für Menschzentrierte Künstliche Intelligenz, Universität Augsburg)

 

Kurzbiographie

Dr. Yekta Said Can received the B.Sc., M.Sc., and Ph.D. degrees from Bogaziçi University, Istanbul, Turkey, in 2012, 2014, and 2020, respectively. He has worked as a Teaching Assistant at Bogaziçi University for six years during his Ph.D. After obtaining his Ph.D. degree, he worked as a Postdoctoral Researcher in a European Union’s Horizon 2020 ERC project (UrbanOccupations) for applying computer vision techniques to retrieve information from historical documents for two years. He is currently working on recognizing emotions and stress at the University of Augsburg as a Postdoctoral Researcher. His research interests include biometrics, document analysis, physiological signal processing, affective and wearable computing, and machine learning.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

A common theme of many science-fiction scenarios is that the physician of the future seems to be able to wirelessly scan a person and determine his or her health status without physical interference. A common theme of today’s diagnostics is the use of invasive procedures, stationary devices, wires, adhesive-based connections, and the associated limitations in flexibility and comfort. At the same time, the development of sensing modalities for unobtrusive sensing is ever increasing, both in the scientific community as well as in the industry. In part, this ever-accelerating development is fueled by the developments in artificial intelligence. In this talk, we will explore how artificial intelligence, in particular machine learning, sensor fusion, and computer vision may be used to extract diagnostic information using cameras and other types of unobtrusive sensors.

 

Referent: Prof. Dr.-Ing. Christoph Hoog Antink (Lehrstuhl für Künstlich Intelligente Systeme der Medizin, TU Darmstadt)

 

Kurzbiographie

Prof. Christoph Hoog Antink received the M.Sc. in mechanical engineering from the University at Buffalo, USA, in 2011, and the Dipl. Ing. and Dr. Ing. (Ph.D.) degrees in electrical engineering from RWTH Aachen University, Germany, in 2012 and 2018, respectively. Until the end of 2020, he was Head of the Medical Signal Processing Group at the Chair for Medical Information Technology, RWTH Aachen.

Since then, Christoph Hoog Antink is Assistant Professor (tenure track) and Head of KIS*MED (AI Systems in Medicine Laboratory), TU Darmstadt, Germany. His research interests include unobtrusive sensing of vital signs, sensor fusion, and machine learning in medicine.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

Efforts made to make data findable, accessible, interoperable and reusable (FAIR) have shown to lead to a more sophisticated data stewardship in various scientific disciplines. One positive consequence of a so-called FAIRification process is the increase in data use and reuse, another observed benefit is the reduction of irreproducle scientific results due to more transparency and richer documentation.

In this talk, I will introduce the FAIR principles in detail using examples from our own work at the Medical Informatics Lab. We conducted a baseline assessment of the NFDI4Health metadata schema and worked to FAIRify the core data set used for diabetes research in Germany, the SHIP laboratory data set, and biosimulation models. Currently we run a FAIR assessment of the Medical Informatics Core Data Set, funded by EOSC. I will highlight the usefulness of making own data FAIR and present a few tools to do so. I will also discuss with you why it is not easy to achieve FAIRness, and we will learn about possible next steps to implement the FAIR principles in the German medical research community at large.

For example, we recommend that key performance indicators be established to adequately evaluate how data FAIRness benefits health research. Future work also involves developing guidelines for researchers in the form of a simple checklist they can follow. You can read more about these topics in our recent scoping review on FAIR data in health care research (https://www.jmir.org/2023/1/e45013).

 

Referentin: Prof. Dr.-Ing. Dagmar Waltemath (Lehrstuhl für Medizinische Informatik und Leiterin des Core Unit Data Integration Center, Universitätsmedizin Greifswald)

 

Kurzbiographie

Prof. Dagmar Waltemath is a computer scientist with a specialisation on database and information systems. Since December 2018, she holds a professorship of Medical Informatics at the University Medicine in Greifswald, Germany (Institute of Community Medicine). Since May 2020, she leads the Research Data Integration Center at the University Medicine. The Core Unit Data Integration Center (CU DIZ) supports clinician scientists in all aspects of research data management and develops methods and tools to extract primary care data from different clinical information systems into a standardised, accessible research data infrastructure. They work on concepts to convert clinical data silos into an integrated data pool that can be accessed by clinicians through a data transfer unit.

Dagmar Waltemaths research is focused on data management in biomedicine, data integration across health care providers and provenance of clinical research data items within clinical information systems. Furthermore, she develops methods and tools for model management in computational biology and is actively involved in COMBINE standardisation efforts.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

Deep convolutional neural networks (DCNNs) can be trained to recognize and classify rare genetic disorders based on their characteristic facial features. GestaltMatcher is such an AI that learned to construct a clinical face phenotype space (CFPS) in which portrait images can be placed. Distances in between medical images in the CFPS can be used to quantify syndromic similarities. Cluster analysis in that space reveals close proximity of molecular pathway diseases and can, therefore, be used to support "lumping decisions". Furthermore, the pleiotropic genes causing multiple disorders can also result in distinct facial gestalt. Thus, cluster analysis can also provide evidence for "splitting decisions", that is indication for different pathomechanisms based on gestalt analysis. 

 

Referent: Prof. Dr. med. Peter Krawitz (Institut für Genomische Statistik und Bioinformatik, Universitätsklinikum Bonn)

 

Kurzbiographie

Prof. Peter Krawitz studied Medicine and Physics in Munich. He continued his specialization in Medical Genetics at Charité Berlin and did a postdoc in Bioinformatics. As a clinician scientist, he was able to identify the disease-causing variants in the gene PIGV in patients with Hyperphosphatasia with Mental Retardation (aka Mabry syndrome) by exome sequencing in 2009. Peter Krawitz contributed to establishing next-generation sequencing protocols in routine health care with a research focus on deep learning methods to analyze medical imaging data.

In 2017 he was appointed a full professor at the University of Bonn and established the Institute for Genomic Statistics and Bioinformatics (IGSB).

Veranstaltungsort: Großer Hörsaal (2.OG, Raum 047, Universitätsklinikum)

 

Abstract

Clinical decision support tools with artificial intelligence (AI) technologies could synthesize the abundance of data collected in hospitals to help clinicians in identifying which patients may need specific attention and steer additional mental health resources to those at high risk. However, methodological and implementation challenges, such as algorithmic fairness and generalizability, deserve further attention when developing AI-tools within the (mental) healthcare domain. In this talk, we will address the potential and pitfalls for developing responsible AI for mental healthcare.

 

Referentin: Assistant Professor Anne de Hond (Abteilung für Medizinische Statistik und Bioinformatik, Universitätsklinikum Utrecht)

 

Kurzbiographie

Anne de Hond is an assistant professor in the data science program of the University Medical Center Utrecht. She completed a master's degree in econometrics and worked as a data scientist and researcher in the AI-team at the Leiden University Medical Center during her PhD studies, where she developed and implemented several AI algorithms. Her research focuses on the responsible development, validation, and implementation of AI algorithms in healthcare. Amongst other applications, she collaborated with Stanford University to develop an AI model to predict depression risk in cancer patients. Her current research interests include validation methods for clinical AI, including large language models like ChatGPT, as well as explainability and fairness of AI algorithms.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

Mobile out-of-the-lab monitoring of electrophysiology data is increasingly in the focus for applications like sports science, brain computer interfacing, neurofeedback, or continuous self-optimization. In this talk, we will discuss technological and methodological requirements as well as selected solutions of state-of-the-art sensors and signal processing methods. We will discuss the impact of mobile recording conditions in the light of the progress of wearable multichannel device development for real-life self-applied use.

 

Referent: Jun.-Prof. Dr.-Ing. Patrique Fiedler (Institut für Biomedizinische Technik und Informatik, TU Ilmenau)

 

Kurzbiographie

Jun.-Prof. Patrique Fiedler studied electrical engineering and information technology at the Technical University of Ilmenau and received his PhD in biomedical engineering in 2017. He then moved to industry from 2017 to 2021 and took on various development, project and product management activities at an internationally active medical technology manufacturer. Additionally, Patrique Fiedler was a visiting scientist at the University of Porto in Portugal and the University of Pescara-Chieti in Italy on several occasions.

Since 2021, Patrique Fiedler is a junior professor and head of the department "Data Analysis in Life Sciences" at the Institute of Biomedical Engineering and Computer Science of the Faculty of Computer Science and Automation at the TU Ilmenau. His research topics include data fusion and analysis of multimodal data sets and body sensor networks, as well as the exploration of novel sensor concepts for biomedical engineering. Another focus is the development of online-capable analysis methods for hardware-related data processing.

Veranstaltungsort: Großer Hörsaal (2.OG, Raum 047, Universitätsklinikum)

 

Abstract

Die Integration von Artificial Intelligence (AI) und Data Science hat das Potenzial, die Anästhesie und Intensivmedizin zu revolutionieren und zur treibenden Innovationskraft in diesem medizinischen Fachbereich zu werden. Der Vortrag wird einen Überblick über diese spannende Entwicklung bieten mit einigen Beispielen aus der eigenen Forschung. Unter anderen werden die folgenden Themen behandelt:

 

• Geschichte der medizinischen Daten - Kurzer Überblick über die Entwicklung von Data Science in Anästhesie und Intensivmedizin (A&I).

• Erörterung, wie AI-Techniken im Feld der A&I genutzt werden können, um große Mengen von medizinischen Daten zu analysieren, Muster zu erkennen und innovative Forschungsansätze zu entwickeln.

• Beispiele verschiedener Anwendungsbereiche, einschließlich Decision Support Systeme, Überwachung und Vorhersage von Vitalparametern (Schwerpunkt auch perioperatives Monitoring!) und Artefakterkennung.

• Insbesondere auf das Reinforcement Learning soll im Detail eingegangen werden als leistungsstarke AI-Technik, die für die Optimierung von Behandlungsplänen in A&I eingesetzt werden kann.

• Schließlich soll auf das Konzept von Explainable AI und Computer Obedience näher eingegangen werden - da Vertrauen in AI-Systeme vs. kritische Betrachtung von AI-Systemen auch von großer praktischer Bedeutung ist.

 

Referent: Univ.-Prof. Dr. Oliver Kimberger (Abteilung für Allgemeine Anästhesie und Intensivmedizin, Medizinische Universität Wien)

 

Kurzbiographie

Prof. Oliver Kimberger is a postdoc consultant anaesthesiologist. He completed a postgraduate course in Statistics and Biometrics at the University of Heidelberg (MSc) and an MBA in Health Care Management at MedUni Vienna. He also spent a two-year study placement at the Inselspital University Hospital in Berne (Switzerland).

Kimberger is the director of trauma anaesthesia and works as Deputy Head of the Division of General Anaesthesia and Intensive Care Medicine of the Medical University of Vienna.

At the Medical University of Vienna, he is also the director of the department’s Data Science group, which is attached to both the Division of General Anaesthesia and Intensive Care Medicine and the Ludwig Boltzmann Institute for Digital Health and Patient Safety which is hosted by the Division of Anaesthesia and Intensive Care Medicine of the Department of Anesthesia, Critical Care and Pain Medicine of MedUni Vienna and Vienna General Hospital.

He is a member of the Austrian Society of Anaesthesiology, Resuscitation and Emergency Medicine and of the European Society of Anaesthesiology. His research interests include topics such as microcirculation & fluid management, patient temperature management, Big Data, Artificial Intelligence and the digital transformation of medicine.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

 Epigenetic mechanisms play a crucial role in establishing and preserving cellular states and function throughout an organism’s life. DNA methylation is an essential part of this multilayered regulation and displays a highly conserved, characteristic bimodal distribution across most somatic cell types. For yet unknown reasons, the distribution of DNA methylation is fundamentally different tumors, an intermediate gain of DNA methylation at certain CpG islands and a partial loss of methylation across gene-poor regions. These two features make DNA methylation an ideal readout for tumor-type prediction from sequencing data.

Specifically in the field of neuropathology, DNA methylation-based tumor type classification has recently become part of the WHO. However, even though the field aimed to achieve an intraoperative differential diagnosis for decades, accomplishing this within a clinically relevant timeframe has remained elusive.

Recent advances in third-generation sequencing technologies have brought this goal within reach. To allow for intraoperative CNS tumor type classification within less than one hour, we developed MethyLYZR, a Naïve Bayesian framework enabling live classification of cancer epigenomes using fully tractable single-CpG resolution modeling while avoiding the need for feature selection or ad hoc model training. MethyLYZR can be run in parallel to an ongoing Nanopore experiment with negligible computational cost and provides clinically relevant and accurate cancer classification results within 15 minutes of sequencing (94% accuracy). 

 

Referentin: Dr. Helene Kretzmer (Abteilung für Genomregulation, Max-Planck-Institut für molekulare Genetik Berlin)

 

Kurzbiographie

Dr. Helene Kretzmer did her PhD in Bioinformatics at the University of Leipzig, where she focused on the development of methods for DNA methylation sequencing analysis and their application on cancer data in the framework of the International Cancer Genome Consortium. Her main interest is the establishment and stabilization of cell fates. Currently, she is studying how epigenetic factors regulate early mammalian embryonic development using scRNAseq and WGBS data and is also analyzing the epigenetic footprint of cancer to better understand the underlying molecular mechanisms. Helene Kretzmer joined Alex's group in late 2017.

Veranstaltungsort: Großer Hörsaal (2.OG, Raum 047, Universitätsklinikum)

 

Abstract

Spatial transcriptomics technologies have revolutionised how we characterise cellular systems - they are able to profile gene expression in the native tissue context, thus preserving cellular organisation in tissue and niches. In this talk I will present some of the challenges associated with analysis spatial transcriptomics data, in particular cell segmentation, and how we are trying to develop tools to address these issues.

 

Referent: Dr. Naveed Ishaque (Gruppenleiter für Computergestützte Onkologie, Berlin Institute of Health and Charité)

 

Kurzbiographie

Dr. Naveed Ishaque obtained a M.Sc. in Bioinformatics and Systems Biology at Imperial College London and undertook his Ph.D. under Jonathan Jones at Sainsbury Laboratory in Norwich, focusing on the genetic basis of plant-pathogen interactions. As for his post doc, he joined the Heidelberg Centre for Personalised Medicine under Benedikt Brors and Roland Eils in 2013. In 2018, he moved with Roland Eils to Berlin where he leads the Computational Oncology research group at the Berlin Institute of Health. His group develops and apply bioinformatics methods for analysing omics data for human diseases, with a strong interest in spatial omics.

Veranstaltungsort: Hörsaal N2045 (Fakultät für Angewandte Informatik)

 

Abstract

Electrical sensors are used for many neurological or cardiological examinations. Here this is referred to as electroencephalography (EEG) or electrocardiography (ECG). For this purpose, electrodes are placed over the chest or on the head and electrical voltages are measured. This works quite well, but the propagation from the sources (the heart or the brain) to the sensors usually follows very "unusual" (material-dependent) paths. Alternatively, newly developed magnetic sensors can be used - here propagation takes place in a way that is noticeably more "material-independent" for most people. Due to the magnetic measurements, it might be possible in the future to perform much more accurate medical analyses, even without body contact - similar to the "tricorder" from "Star Trek".

In this talk, the Collaborative Research Center 1261 (Magnetoelectric Sensors: From Composite Materials to Biomagnetic Diagnostics), which deals with the research of novel magnetoelectric sensor systems, will be presented first. Then, some of the sensor principles designed to measure low-frequency, extremely weak magnetic fields in typical environments, i.e., without magnetic shielding and without expensive cooling systems, are presented. A brief outlook on biomedical applications for these types of sensors concludes the talk.

 

Referent: Prof. Dr.-Ing. Gerhard Schmidt (Institut für Elektrotechnik und Informationstechnik, Christian-Albrechts-Universität zu Kiel)

 

Kurzbiographie

Prof. Gerhard Schmidt received his diploma in electrical engineering from the Technical University of Darmstadt in 1996 and his Ph.D. on signal processing in 2001. He then worked as a researcher and group leader at Harman/Becker Automotive Systems in Ulm. From 2009 to 2010 he was head of the "Acoustic Speech Enhancement" department of SVOX in Ulm, Germany, as well as Industry Professor (part-time W3) at the Technical University of Darmstadt.

Since 2010, he has been Professor of Digital Signal Processing and Systems Theory at Christian Albrechts University in Kiel. His research interests include speech and audio processing, medical signal processing, and signal processing for underwater applications.

Veranstaltungsort: Großer Hörsaal (2.OG, Raum 047, Universitätsklinikum)

 

Abstract

Translational informatics plays a crucial role in transforming biomedical research, with a special emphasis on the sharing and reuse of data. This talk will present an overview of informatics research at the Berlin Institute of Health at Charité - Universitätsmedizin Berlin. It will proceed to outline Charité's local strategies and expand to address regional and national infrastructures and platforms in Germany.

 

Referent: Univ.-Prof. Dr. Fabian Prasser (Leiter AG Medizininformatik, Charité Universitätsmedizin Berlin)

 

Kurzbiographie

Prof. Fabian Prasser was born in 1982 in Starnberg and studied computer science with a minor in theoretical medicine at the Technical University of Munich. He completed his interdisciplinary doctoral thesis both at the Institute for Medical Informatics, Statistics and Epidemiology and in computer science at the Chair for Database Systems. In doing so, he dealt with modern methods of data integration for translational medical research. Another focus of his work is data protection, which he also examined in his habilitation thesis. Fabian Prasser received several awards for his work on data anonymisation, including the Johann Peter Süßmilch Medal of the German Society for Medical Informatics, Biometry and Epidemiology in 2017. Until his move to the Charité, Fabian Prasser was technical coordinator of the consortium "Data Integration for Future Medicine (DIFUTURE)" in the medical informatics initiative of the Federal Ministry of Education and Research.