Keynote Lectures

Abstract
The emergence of generative pre-trained transformers based on large sets of natural language text training data has triggered an abundance of applications ranging from chatbots, essay writing, poem generation, to text mining. However, at the moment, there is little scientific evidence on the way these AI tools can be integrated into large software applications. AI engineering is the domain of software engineering concerned with the architecture and development of these applications.

In this keynote, we will present and discuss the Care2Report research program of Utrecht University that aims to design generic architectures of software applications for the automated reporting of human activity. Innovative interaction and reasoning are now becoming available using off-the-shelf AI technologies: generative pre-trained transformers, large language models, speech recognition, action recognition, ontologies, knowledge graph databases, agentic frameworks, and several more. We apply this general vision in the healthcare domain due to the societal need in this domain: high administrative burden where administrative duties are reported to take 20 to even 40% of the working time.

We will highlight the design principles and the experimentation of the research program. We show how networks of architectural pipelines are being deployed to configure the AI technologies into one overall application for the reporting of medical consultations. Prompt engineering plays a major role in the semantic interpretation of the natural language interaction for the summarization tasks. The application in the healthcare domain requires proper recognition of anatomic elements, symptoms, observations, diagnosis, and treatment policies. This recognition is configured based on a so-called medical guideline ontology derived from the publicly available guidelines of healthcare professionals. We discuss how these technologies can be applied in similar arrangements of the domain of police reporting and social care of municipalities.

We end with an outlook of the future of this exciting, yet challenging, research endeavor.

References:
van Zandvoort, D., Wiersema, L., Huibers, T., van Dulmen, S. and Brinkkemper, S. (2024). Enhancing Summarization Performance Through Transformer-Based Prompt Engineering in Automated Medical Reporting. In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies - HEALTHINF; ISBN 978-989-758-688-0; ISSN 2184-4305, SciTePress, pages 154-165. DOI: 10.5220/0012422600003657

Faber, W., Bootsma, R., Huibers, T., van Dulmen, S. and Brinkkemper, S. (2024). Comparative Experimentation of Accuracy Metrics in Automated Medical Reporting: The Case of Otitis Consultations. In Proceedings of the 17th International Joint Conference on Biomedical Engineering Systems and Technologies - HEALTHINF; ISBN 978-989-758-688-0; ISSN 2184-4305, SciTePress, pages 585-594. DOI: 10.5220/0012422300003657

Maas, L., Geurtsen, M., Nouwt, F., Schouten, S. F., Van De Water, R., Van Dulmen, S., ... & Brinkkemper, S. (2020, January). The Care2Report System: Automated Medical Reporting as an Integrated Solution to Reduce Administrative Burden in Healthcare. In HICSS (pp. 1-10).

Abstract
In this talk, I will take the audience on a journey through my 25-year research career, spanning multiple continents, disciplines, and academic roles. Starting in Barcelona (Catalonia), my path led to international collaborations in Grenoble (France), Tokyo (Japan), Cambridge (UK), Paris (France), Tianjin (China), and beyond, exploring the boundaries between theory and application in biomedical signal processing. My early research focused on Blind Source Separation (BSS), Independent Component Analysis (ICA), and their applications to brain signals (EEG, fMRI). Over time, I expanded this work to include indirect measurements of brain activity using speech, handwriting, and gait, while collaborating with leading institutions such as the RIKEN Brain Science Institute (Japan), the Brain Mapping Unit at Cambridge (UK), and Juntendo University (Japan). Currently, my research focuses mainly on biomedical signal processing, including EEG, brain-computer interfaces for neurorehabilitation and identification of neurodegenerative disease markers through voice and handwriting analysis. I will also share recent initiatives in mental health research with companies such as TopDoctors. Combining teaching and research has been a fundamental part of my career. I will discuss strategies for managing both roles, from managing heavy teaching responsibilities to maintaining an active research agenda. Finally, I will outline future directions for my work and the mentoring of new doctoral students. This talk reflects the crossing of multiple boundaries - geographical, disciplinary and academic - as I share how my career has bridged the gap between theoretical research and practical applications.

Abstract
Genome instability and aberrant alterations of transcriptional programs both play important roles in cancer. However, their relationship and relative contribution to tumor evolution and therapy resistance are not well-understood. Single-cell RNA sequencing (scRNA-seq) has the potential to investigate both genetic and non-genetic sources of tumor heterogeneity in a single assay. I will describe how incorporation of genetic information about the haplotype configuration of an individual based on population genetics can yield notable improvements in the ability to detect chromosomal aberrations (CAs) in scRNA-seq data. Such an approach is better able to resolve subclonal populations, helping to delineate contributions of genetic and non-genetic mechanisms in cancer. Somatic mutations also accumulate in normal tissues with age. Although widespread somatic mosaicism in the form of single-nucleotide variants has been well characterized, the extent and consequences of aneuploidy in healthy tissues remain largely unknown.  Extending sensitivity and specificity of our CA detection approach, we carried a large-scale survey of mosaic chromosomal alterations (mCAs) in the Genotype-Tissue Expression (GTEx) project. We show that prevalence and genome-wide patterns of mCAs vary considerably across tissue types, suggesting tissue-specific mutagenic exposure and selection pressures.

Abstract
Radiological images and conventional histopathological tissue sections contain valuable clinical information for prevention, diagnosis, treatment, and prognosis. With the development of image analysis and machine learning techniques, we can extract image phenotypes or image biomarkers of diseases from radiological images and histopathological sections. These image biomarkers can not only help doctors detect and diagnose diseases more accurately from histological images but also attempt to predict patients' recurrence risk, disease invasiveness, patient survival rate, and patient response to treatment. At the Jiangsu Provincial Key Laboratory of Intelligent Medical Image Computing (iMIC), our team is devoted to exploring advanced data- and knowledge-driven tools for extracting and analyzing "image biomarkers" from radiological images and digital tissue samples. In this talk, I will introduce our group's recent work in the field of medical image computing, including 1) prediction of microvascular invasion of hepatocellular carcinoma based on liver CT images; 2) quantitative analysis and survival prediction of tumor microenvironment in patients with cholangiocarcinoma; 3) single-cell morphology and topological maps reveal the diversity of breast cancer ecosystem.