Document Actions

2026-02-23: xGNN4MI

In this study, we introduce xGNN4MI, an open-source framework for graph-based modelling of 12-lead ECG signals. By explicitly encoding spatial relationships between ECG leads and their temporal structure, the framework provides a transparent and reproducible pipeline for GNN-based cardiovascular disease classification.

Excited to share our latest work published in Springer Nature npj Digital Medicine (February 2026). Title: xGNN4MI: Explainability of Graph Neural Networks in 12‑Lead Electrocardiography for Cardiovascular Disease Classification

In this study, we introduce xGNN4MI, an open-source framework for graph-based modelling of 12-lead ECG signals. By explicitly encoding spatial relationships between ECG leads and their temporal structure, the framework provides a transparent and reproducible pipeline for GNN-based cardiovascular disease classification.

Key Highlights:
-         Integrated explainability workflow
-         physiologically meaningful insights
-         external validation on population-based cohort
-         open-source and reproducible


📖 Read the full paper: https://lnkd.in/eeyxrdmh

Big thanks to to my co-authors and collaborators at Department of Medical Informatics, University Medical Center, CIDAS University of Göttingen, Department of Artificial Intelligence in Biomedical Engineering, University Erlangen-Nürnberg, Institute of Bioinformatics, University Medicine Greifswald, Department of Health Technology, Technical University of Denmark, PredLmed Gießen for their valuable contributions. 🙏

Miriam Cindy Maurer, Philip Hempel, Kristin Steinhaus, Hryhorii Chereda, Marcus Vollmer, Dagmar Krefting, Nicolai Spicher and Anne-Christin Hauschild
 
Bild aus Publikation xGNN4MI