El Instituto Andaluz en Data Science and Computational Intelligence, DaSCI, anuncia la organización del Workshop gratuito «Health and AI», sobre la cada vez mayor importancia que la Inteligencia Artificial, IA, juega en la Salud. Esta actividad se enmarca dentro de las previstas en el proyecto Inteligencia Artificial, Ética, Responsable y de Propósito General (IAFER).
Se realizará el día 18 de Febrero de forma presencial en el Edificio UGR-AI en el PTS, y de forma online en el enlace indicado en la web del evento.
Solo los primeros 50 inscritos podrán asistir de forma presencial, el resto podrá acceder de forma online. Los ponentes son especialistas en el uso de IA para problemas médicos complejos de las Universidades de Granada, Nottingham, y el Imperial College London.
Programa de charlas previstas:
10:00-10:30 Responsible Artificial Intelligence for the Afterlife
Speaker: Elvira Perez Vallejos, University of Nottingham
Abstract: Artificial intelligence is revolutionising how we interact with the world, and death is no exception. During this talk, Prof. Perez will present the UKRI research programme RAi UK [[https://rai.ac.uk/]] , a £35M investment developed to deliver world-leading best practices for how to design, evaluate, regulate, and operate AI-systems in ways that benefit people and society. The talk then explores a specific case study: DeathTech, (i.e., emerging technologies that aim to preserve the memories of our loved ones digitally). From creating conversational avatars to reconstructing voices and faces using generative AI, the possibilities seem limitless. However, it’s crucial to approach this with responsibility. The talk examines how AI can help us cope with grief and keep the memories of our loved ones alive, while emphasising the need for a robust ethical framework to ensure these technologies are used for the benefit of society.
10:30-11:00 Subgrouping Germinal Center-Derived B-Cell Lymphomas based on Machine Learning-deduced DNA Methylation Modules
Speaker: Coral del Val Muñoz, University of Granada
Abstract: Existing subgrouping techniques for diffuse large B-cell lymphomas based on morphology, transcriptomics, or genetic alterations are hindered by overlapping molecular signatures, intratumoral heterogeneity, and inconsistent reproducibility. Although DNA methylation profiling has successfully stratified solid tumors and leukemias, its application to mature B-cell lymphomas (FL and DLBCL) is challenged by a continuous rather than discrete distribution of methylation states. To address this, we use an unsupervised framework that integrates preliminary DNA methylation clustering with fuzzy non-negative matrix factorization (FNMF) to extract robust, interpretable methylation markers. This approach is able to capture inherent ambiguities in methylation patterns, thereby enhancing subtype delineation. Results revealed 300 CpGs forming four methylation modules, which ordered the lymphomas into seven methylation patterns (MP1-7). These MP1-7 showed significant associations with biological features of the lymphomas and were replicated in external samples.
11:30-12:00 Enhancing the analysis of Idiopathic Pulmonary Fibrosis through Longitudinal FVC Trajectories and Endotype Identification Using Advanced Computational Techniques
Speaker: Hernan Fainberg, Imperial College London
Abstract: Idiopathic pulmonary fibrosis (IPF) patient clusters were identified through sensitivity analysis enhanced machine learning (ML) models. Two studies were conducted: 1) analyzing FVC trajectories and 2) classifying endotypes. In FVC analysis, RF models trained with MCMC simulations showed the lowest NRMSD. Four FVC clusters were obtained, each associated with distinct mortality risks. A significant FVC decline in the first-year post-diagnosis indicated higher mortality risk. For endotype classification, three clusters – Basement Membrane, Epithelial Injury, and Crosslinked Fibrin – were identified, each associated with unique survival rates and biomarker profiles. Sensitivity analysis and replication in an independent dataset confirmed these results. Conclusions: Sensitivity analysis enhances ML models by improving their robustness, interpretability, and generalisability in analysing longitudinal FVC trajectories and endotype classification in IPF. These findings support distinct lung function trajectories and endotypes, each with unique clinical characteristics, and offer insights into IPF progression for future clinical trials and improved patient management.
12:00-12:30 Multi-Species and Multi-Antibiotic Resistance Classification with MALDI-TOF
Speaker: Daniel Peralta Cámara, University of Granada
Abstract: Antimicrobial Resistance (AMR) poses a significant global health threat, impacting clinical treatments, agriculture, and public health. Although mass spectrometry techniques like MALDI-TOF provide opportunity for rapid AMR detection, state-of-the-art models are largely based on manual analysis or classic machine learning techniques that are heavily dependent on preprocessing. In this talk, we will discuss MSDeepAMR, a convolutional network that predicts AMR from the raw mass spectrometry data. The model was tested against several antibiotics and bacterium species, highlighting the accuracy variability in each case. The model was also tested under transfer learning conditions, to estimate its performance when dealing with data from small hospitals. Furthermore, we discuss an adaptation of MSDeepAMR to incorporate multi-species and multi-antibiotic classification, enabling simultaneous prediction of resistance across multiple antibiotics and bacterial species. Results demonstrate enhanced prediction accuracy and generalizability, highlighting the potential of integrating multi-label classification with MALDI-TOF data for efficient and scalable multidrug AMR detection.
- Fecha: martes 18 de febrero de 2025
- Lugar: Edificio UGR-AI
- Horario: de 10-12:30
- Organiza: Instituto Andaluz en Data Science and Computational Intelligence, DaSCI
- Más información: web del evento