Bio: Ninon Burgos is CNRS research director at the Paris Brain Institute, co-head of the ARAMIS Lab, and a fellow of PR[AI]RIE, the PaRis Artificial Intelligence Research InstitutE. She completed her PhD in 2016 at University College London, and obtained her Habilitation (HDR) from Sorbonne Université in 2022. In 2019, she received the ERCIM Cor Baayen Young Researcher Award.

Her research focuses on the processing and analysis of medical images, the use of images to guide diagnosis, and the application of these methods to the clinic. In particular, she has contributed to: i) anomaly detection using traditional image processing techniques and deep generative models, ii) the translation of machine learning approaches for quality control and computer-aided diagnosis into clinical practice, iii) reproducible medical image processing and computer-aided diagnosis with machine learning, and iv) the development of open-source software.

Abstract: I will cover advances in unsupervised anomaly detection for neuroimaging using generative models. I will present a comprehensive evaluation of variational autoencoders for pseudo-healthy image reconstruction and anomaly detection in brain FDG PET, and will introduce Bayesian flow networks, a new class of generative models for this task. Finally, I will discuss proof-of-concept results for detecting white matter hyperintensities in heterogeneous brain clinical MRI data.

Bio: Martin Krallinger is the head of the NLP for Biomedical Information Analysis (NLP4BIA) team at the Barcelona Supercomputing Center (BSC) and an expert in biomedical and clinical language technologies. His work focuses on developing language technologies for health-related applications, including rare diseases, drug safety, biomaterials, cardiovascular diseases, toxicology, and occupational health, with a particular emphasis on the development and evaluation of multilingual solutions. He is especially known for advancing the benchmarking and evaluation of biomedical NLP and LLM-based tools, and for organizing more than 25 shared tasks as part of international community challenges such as BioCreative, BioASQ, BIONLP-ST, IberEval, IberLEF, biomedical WMT, and eHealth CLEF. His team has made major contributions to high-quality annotated datasets, medical corpora, and annotation protocols, supporting the development of state-of-the-art transformer-based NLP systems using BSC’s computational resources, and resulting in collaborations with major hospitals in Spain and across Europe.

Abstract: Biomedical natural language processing (NLP) has advanced rapidly with the emergence of large language models (LLMs), creating new opportunities to extract and leverage information from unstructured data sources such as clinical records, scientific literature, and social media. Despite this progress, significant challenges remain, particularly in the creation of high-quality annotated corpora, the development of reliable named entity recognition (NER) systems, and the establishment of robust evaluation frameworks. This talk will highlight recent strategies in biomedical and clinical NLP, with a focus on multilingual resources that extend capabilities beyond English to languages such as Spanish, Catalan, Italian, Dutch, and Swedish. It will present practical applications in areas including rare diseases, cardiology, predictive modeling along with an overview of community-driven shared tasks and evaluation initiatives that are shaping progress in the field.

Bio: Guillem Anglada-Escudé obtained his Phd in Astrophysics by the Universtat de Barcelona in 2007. His work focus has been on the development of methods and data analysis techniques for the detection of extrasolar planets, and has worked at various academic institutions in the US, Germany, UK and Spain. He led various studies leading the discovery of several nearby exoplanets to the Solar System, including Proxima b, the neares exoplanet to the Sun. In the recent years he started working on Machine Learning and the use of AI related tools for data analysis in astronomy. His is a staff scientist at the Insitute de Ciencies del Espai/CSIC. Since 2025, he is a Seconded National Expert at the European Research Council Executive Agency, where he applies data science and Machine Learning techniques to support various tasks within the agency in Brussels.

Abstract: Machine learning and (more recently) generative AI have been penetrating all fields of science, especially those that are data intensive from the point of view of size, but also their complexity and multimodality. In the first part we will discuss the main results of the recently published report As a result of the study “ERC Frontier Research for AI in Health”, covering from global trends to a deep dive into a handful of ERC funded projects illustrating the unprecedented capabilities enabled by these technologies. Ethical issues and prospects for future use of AI in health drawn from this report will also be presented.

The ERC is progressively deploying AI-based tools in several activities, such as portfolio analyses or quality control tasks related to project funding. This includes classification using predefined or driven taxonomies, use of AI elements to distil specific information from projects (science topics, methodologies), and large-scale classification and quality control checks. This is done by combining LLM, algorithmic methods, and human supervision at design level. It is important to remark that assessment of scientific quality and merit remains an exclusive competence of expert reviewers. We will review practical considerations in the implementation of AI tools to projects and proposals processing. We will provide some examples with published use cases.

Bio: Jordi Abante is a Ramón y Cajal Fellow at the Universitat de Barcelona (Dept. of Biomedical Sciences), where he leads a research group developing biologically grounded machine-learning methods to extract interpretable insights from complex neuroscience data. He trained in Electrical and Computer Engineering and Applied Mathematics (UPC, Texas A&M University, Johns Hopkins University) and completed postdoctoral research at Stanford University and the Universitat de Barcelona.

Abstract: Generative artificial intelligence is rapidly reshaping Health Sciences by enabling models that can learn directly from the structure and complexity of biomedical data. In this talk, I will present three complementary research directions from our group. First, I will discuss our work on genotype-to-phenotype prediction, where we leverage deep generative models and a large genomic model to discover genetic modifiers in a neurodegenerative disease. Second, I will cover our recent work on deep conditional generative models for neural data integration, focused on calcium imaging traces. Finally, I will introduce our ongoing work on a foundation model for neural photostimulation, designed to learn transferable representations and accelerate closed-loop, data-driven interventions. Together, these projects advance a central goal for AI in the health sciences: building generative models that are biologically grounded, scalable, and capable of translating complex biomedical data into actionable insight.

Bio: Agustín Gutiérrez-Gálvez has been an Associate Professor in the Department of Electronics and Biomedical Engineering at the University of Barcelona (UB) since 2017. He holds degrees in Physical Sciences and Electronic Engineering from the University of Barcelona, earned his PhD in Computer Science from Texas A&M University, and completed his postdoc as a Marie Curie fellow at UB.

Abstract: The Signal and Information Processing for Sensing Systems group at IBEC/UB is dedicated to developing computational tools and machine learning workflows aimed at analyzing complex omics data collected from a variety of instrumental plaftorms, including chemical sensor arrays, NMR, GC-IMS, and mass spectrometry. From 2020 to 2025, we have achieved significant milestones, including the creation of the AlpsNMR and GCIMS R packages. These tools have gained widespread popularity, with the AlpsNMR as the most downloaded Bioconductor tool for untargeted 1H-NMR analysis, while the GCIMS package is considered the leading workflow for GC-IMS ion detection. 

Our collaborations with clinical partners, such as Hospital del Mar, Hospital Clínic de Barcelona, and Vall d'Hebron, have yielded encouraging results. We've explored breath analysis for detecting Pseudomonas, identified plasma biomarkers for complications following colorectal cancer surgeries, and developed predictive models for ventilatory failure in patients with COVID-19 and ARDS. We've also made contributions in assessing patient prognosis based on their immunological profiles.

Looking ahead, we are aiming to enhance our capabilities in raw signal preprocessing through techniques such as tensorial deconvolution and better peak alignment. We're also focused on addressing data shift challenges, which often arise in multicenter studies due to batch effects and instrumental variability. Our future work will make use of domain adaptation and orthogonal projections, along with integrating diverse data types, to continue advancing biomarker discovery and improving model interpretability.

About: Ephion Health is a Barcelona-based deep-tech startup transforming patient monitoring through AI-powered digital biomarkers. Our hardware-agnostic platform unifies multimodal data from any set of third-party wearables into a single, objective health score that surpasses the sensitivity of traditional clinical scales. We offer this platform to healthcare providers as a clinical decision support tool, and to pharmaceutical companies as a tool to prove the efficacy of new drugs in clinical trials. Ephion is bridging the gap between raw wearable data and actionable medical insights to accelerate the future of personalized, data-driven healthcare.

Abstract: Current clinical assessments of patient mobility are often subjective and lack granularity. While the integration of wearable sensors offers high-dimensional, continuous data, it introduces a significant analytical challenge: deal with a high number of highly correlated spatiotemporal and kinematic parameters. Here, we explore how AI can turn this high-dimensional measurments into robust, scalable, reproducible and clinically meaningful biomarkers and how these methods can be flexibly configured to address different clinical and analytical goals.