Unlocking the Potential of Health Data: Navigating the Challenges of Predictive Modeling in Clinical Decision-Making (XtraTrack Season 2026)

20-04-2026

The adoption of AI systems in medicine has become an essential asset for analyzing the vast amounts of complex multimodal health-related data generated daily. However, several challenges persist, including limited sample sizes, systematic biases, and the lack of interpretable and reliable predictions. To address these challenges, novel machine learning and bioinformatics algorithms are essential to support the analysis of the emerging complex multimodal biomedical data and their transition toward clinical applications. The FAIrPaCT project develops federated AI to allow the analysis of heterogeneous clinical, molecular, and medical imaging data from pancreatic cancer patients across multiple German institutions without sharing data. In SATL and MGPath, transfer learning is employed to overcome data scarcity and heterogeneity. SATL is facilitating the translation of research findings between human and model organisms in biomedical research. MGPATH is a parameter-efficient vision–language framework designed for improving cancer subtypes classification in scarce datasets. As the demand for reliable and trustworthy AI systems in healthcare continues to grow, it has become increasingly clear that the current limitations of explainable AI must be addressed, particularly the issues of robustness, comparability, and human oversight. Novel developments such as BenchXAI, xGNN4MI or Clarus aim to address these challenges. BenchXAI, a comprehensive evaluation framework, serves to systematically benchmark XAI methods across multiple biomedical data types, additionally facilitating an ensemble integration across different XAI methods and allowing for robust majority vote explanations. xGNN4MI integrates spatial and temporal information via Graph Neural Networks and XAI to support the interpretation of 12-lead ECGs for cardiovascular disease classification. Finally, the interactive XAI platform CLARUS aims to bridge the gap between AI researchers and domain experts. These contributions pave the way for more robust, trustworthy, and clinically relevant AI integrating routine multimodal health data to advance personalized medicine and improve clinical decision-making, ultimately improving patient outcomes.

Prof. Dr. Anne-Christin Hauschild - : Institut for Predictive Deep Learning for Medicine and Healthcare, Justus-Liebig University Gießen

Prof. Dr. Anne-Christin Hauschild (Dr. rer. nat.) heads the Institute for Predictive Deep Learning for Medicine and Healthcare at Justus-Liebig University Gießen. Previously she was a junior professor at the Institute of Medical Informatics at the University Medical Center Göttingen and headed the Clinical Decision Support group. Her research is dedicated to developing machine learning methods and systems medicine approaches for analyzing biomedical data and their transfer into clinical practice. Her group focuses on integrative analysis of multi-modal medical data, such as electronic patient records, molecular data such as genotypes, gene and protein expression, and medical imaging data to support disease prediction and therapy optimization. ML technologies apply successfully in numerous health research domains, such as oncology, psychiatry, and cardiology. However, several challenges persist, impeding the translation of these advances into research and practice. Specifically, limited sample sizes, data privacy, and systematic biases within individual patient cohorts contribute to data scarcity and heterogeneity in medical registries and biomedical data. Additionally, the lack of interpretable and reliable predictions undermines trust in otherwise highly accurate models. Prof. Hauschilds group aims to address these challenges through employing and developing novel computational architectures and algorithms, including foundation models and transfer learning to overcome data integration issues and accommodate small sample sizes, online and time-critical event prediction. Specialized federated learning enable integrating heterogeneous distributed medical datasets and databases. Finally, the group develops and employs explainable artificial intelligence methods to enhance model and prediction interpretability, particularly aiming to support the development of trustworthy systems for medicine and healthcare.

Organized by GMDS AG MoCoMed / Impressum / Privacy