Data4SmartHealth 2025
27 November 2025, 09:00 – 13:30, Room NOI B1.1.21
Faculty of Engineering, Free University of Bozen-Bolzano, Italy
Talk abstracts
Markus Galler, MD, M.Sc.
Department of Nuclear Medicine, Charité—Universitätsmedizin Berlin
Corporate Member of Freie Universität Berlin and Humboldt—Universität zu Berlin
Artificial Intelligence in PET Imaging: Opportunities and Challenges
The era of artificial intelligence (AI) has introduced numerous opportunities to integrate advanced computational methods into medical imaging. In particular, AI has shown significant potential in enhancing the analysis of positron emission tomography (PET), a key nuclear medicine modality that plays an indispensable role, especially in oncology. Various AI-driven approaches have been developed to support tasks such as tumor segmentation, automated report generation, and the extraction of radiomic features. However, the integration of these technologies also presents substantial challenges. In the medical context, issues such as accountability, data reliability, and the explainability of AI models are critical considerations that must be addressed to ensure safe and effective clinical implementation.
Speakers’ short bio:
Markus Galler is a board-certified physician and researcher in the Department of Nuclear Medicine at Charité – Universitätsmedizin Berlin. With formal training in medicine and physics, he is dedicated to advancing quantitative and computational methods in nuclear medicine. Clinically, he specializes in hybrid imaging and targeted radionuclide therapy for prostate cancer.
(slides)
Andrea Rosani, PhD and Giuseppe di Fatta, PhD
Faculty of Engineering, Free University of Bozen-Bolzano
Image Processing in KNIME: from Classic Image Processing to Artificial Intelligence
The Machine Learning group led by Prof. Giuseppe Di Fatta at the University of Bolzano-Bozen is developing a new extension for KNIME dedicated to advanced image processing. The aim of the project is to integrate a series of nodes dedicated to image processing into the KNIME visual environment, ranging from classic operations — such as filters, transformations and noise reduction techniques — to more advanced components based on artificial intelligence models for automatic image segmentation. In particular, the extension includes nodes that leverage modern architectures such as YOLO (You Only Look Once) and Segment Anything, two cutting-edge approaches to object detection and segmentation. During this talk, we will present a first preview demo of the new extension. It is still a work in progress, with a limited number of nodes already implemented, but it allows us to show the direction of the project and its potential in a concrete way. It will also be an opportunity to gather feedback and suggestions from interested users, so that we can orient development towards the real needs of those who work with image analysis and segmentation in the medical field.
Speaker’s short bio:
Giuseppe Di Fatta has been a full professor at the Faculty of Engineering of the Free University of Bozen-Bolzano since 2022. His teaching and research activities focus on machine learning and data science, with multidisciplinary applications in science and industry.
Previously, he was professor and head of the Department of Computer Science at the University of Reading (UK) from 2016 to 2021. He contributed to the development of the first version of the KNIME platform during his time at the University of Konstanz (Germany) from 2004 to 2006. He also worked at the ICAR Institute of the CNR (Italy) and as a research fellow at the International Computer Science Institute (ICSI) in Berkeley, California.
Andrea Rosani is a researcher at the Faculty of Engineering of the Free University of Bozen-Bolzano. After obtaining his PhD in Information and Communication Technology from the University of Trento, he gained solid experience in image and video processing, with a particular interest in automatic visual content analysis, pattern recognition and artificial intelligence applied to the biomedical and environmental contexts.
(slides)
Mojtaba Hajian, PhD
Faculty of Engineering, Free University of Bozen-Bolzano
Artificial Intelligence and Radiomics for Prostate Cancer Outcome Prediction
In this talk I will present the work performed to date in the AIRFRAME project, which is developing a radiomics pipeline for PSMA PET/CT to predict both treatment response (responder vs. non-responder) and survival / progression-free outcomes in prostate cancer patients undergoing ^177Lu-PSMA radioligand therapy. Accurate predictions are important to avoid ineffective cycles, manage toxicity, and prioritize patients most likely to have benefits. The pipeline starts with segmenting the lesions in whole-body PET scan of Metastatic Castration-Resistant Prostate Cancer (mCRPC), which is performed by medical physics expert. After preparing the lesion masks for all patients, we extract intensity, texture, and shape features at lesion level and consolidate them at patient level, then integrate imaging and key clinical variables to utilize the most informative data for model development. Given the high dimensionality of radiomics versus limited sample sizes, we employ strict dimensionality control and regularization to reduce overfitting and improve generalization. For binary classification, we prioritize calibrated linear/nonlinear models chosen for stability on small and imbalanced datasets. For survival analysis, we use Cox-based and tree-based approaches to model progression- or event-related risk over time, providing clinically interpretable risk analysis beyond a single label.
Speakers’ short bio:
Mojtaba Hajian is a Research Fellow at the Faculty of Engineering, Free University of Bozen-Bolzano, where he works on AI-driven radiomics for prostate cancer within the AIRFRAME project. His research spans machine learning, deep learning, and multimodal medical imaging (especially PSMA PET/CT feature extraction, and predictive modeling for therapy response) developed in collaboration with clinical partners including Paracelsus Medical University (Salzburg) and SABES (Südtiroler Sanitätsbetriebes). He holds a Ph.D. in Information and Communication Technology (ICT) from the University of Calabria and a master’s in biomedical engineering, with prior experience in big-data analytics, OLAP systems, and scalable data pipelines. His recent work integrates advanced feature engineering, explainable AI, and survival analysis to translate imaging biomarkers into clinically actionable insights.
(slides)
Diego Calvanese, PhD
Faculty of Engineering, Free University of Bozen-Bolzano, Italy
Semantic Technologies for Data Access and Integration
In various domains, including medicine, health, and biology, complex data processing tasks in the context of data analytics and machine/deep learning pipelines, require to access and integrate large datasets in a coherent way. Such data are often available only in legacy data sources and highly heterogeneous. Semantic technologies provide allow one to abstract from data representation and concentrate on actual meaning, and have been proposed as a powerful tool to overcome heterogeneity and effectively integrate data. In the talk we present the key elements of semantic technologies for data access and integration, consisting in Knowledge graphs as a uniform data format with the required flexibility, ontologies to represent domain knowledge, and declarative mappings to connect data sources to the knowledge layer. We discuss how these technologies are combined in the Virtual Knowledge Graph approach to data access and integration.
Speakers’ short bio:
Diego Calvanese is a full professor at the Faculty of Engineering of the Free University of Bozen-Bolzano, where he is the head of the Institute of Computer Science and Artificial Intelligence. From November 2019 to October 2024, he has also been a Wallenberg Guest Professor at the Department of Computing Science, Umeå University (Sweden). He received a Ph.D. from Sapienza Università di Roma in 1996. His research interests include knowledge representation and reasoning, virtual knowledge graphs for data access and integration, description logics, Semantic Web, graph data, and data-aware processes modeling and verification. He has been involved in many research projects in the above areas (including FP6 TONES, FP7 ACSI, FP7 Optique, H2020 INODE, HEU CyclOps), acquiring funding for close to 6.4ME. He has authored more than 400 refereed publications, many in the most prestigious international journals and conferences in AI and Databases, with more than 39000 citations and an h-index of 82, according to Google Scholar.
(slides)
Alessandro Mosca, PhD
Laboratory for Applied Ontology (LOA), Institute of Cognitive Sciences and Technologies (ISTC-CNR), Italy
Achieving Semantic Interoperability in Health Records Standards: Who is Doing What, and Where
Starting from an introduction of the notion of semantic interoperability of Electronic Health Records (EHRs), the talk will present a (non-systematic) review of initiatives and projects that have recently proposed semantic technologies-based solutions to overcome the negative effects caused by the proliferation of heterogeneous health information systems and proprietary data models for representing and storing EHR information. The talk itself has been conceived as an opportunity to bring awareness on existing approaches to achieve a semantically consistent exchange of legacy and heterogeneous information among different sectors of a health organization (e.g., clinicians, nurses, labs), and across a health organization boundaries. The most chosen scenarios, technologies, and tools employed to solve interoperability problems, with a special focus on those where ontologies and taxonomies are in use, will be introduced and briefly discussed.
Speakers’ short bio:
Alessandro Mosca is a researcher in the area of Knowledge Representation and Reasoning at the Laboratory for Applied Ontology (LOA) of the Institute of Cognitive Sciences and Technologies (ISTC-CNR). He’s also external member of the KRDB Research Centre and of the Smart Data Factory at the Faculty of Engineering of the Free University of Bolzano. His research interests include the application of mathematical logic in the fields of Artificial Intelligence and Conceptual Modelling for data management and supported decision making. He works on the theoretical and methodological aspects behind the creation of ontology-based data management solutions which, in particular, subsumes the design of formal ontologies, multi-format data integration and access services, efficient ontology-based query answering systems.
(slides)
Francesco Fiz, MD, PhD
Nuclear Medicine Department, Galliera Hospital, Genoa, Italy
The Pillars of Radiomics Research: Sound Scientific Hypotheses and High-Quality Datasets
Radiomics research utilises microscopic details within medical images to identify patterns and link this information to specific outcomes, like patient survival or therapy response. Over the past decade, it has produced numerous scientific papers. However, it remains far from routine clinical application, mainly due to the lack of high-quality, multicentre prospective trials and the ‘black box’ nature of some methods, which generate a radiomics’ signature’ that does not reveal underlying pathological mechanisms. In this presentation, I will outline the fundamentals of radiomics research, including formulating a clear hypothesis before analysis, designing trials with sufficient sample sizes, and interpreting results. Multicentre datasets and strong collaboration across centres are essential for the future of this image analysis method.
Speakers’ short bio: Dr. Francesco Fiz, MD, PhD, is a specialist in Nuclear Medicine with Italian national habilitation for Associate Professor spanning 2018–2029. He earned his PhD in Internal Medicine in 2019, following extensive research in PET/CT image segmentation and radiomics. His work spans institutions in Italy and Germany, with over 30 publications on oncological imaging, including paediatric and liver malignancies. Currently based at Galliera Hospital in Genoa, he leads studies on AI-driven analysis of paediatric cancers.
(slides)
