Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
SOE: Fachverband Physik sozio-ökonomischer Systeme
SOE 10: Networks, From Topology to Dynamics II (joint session SOE/DY)
SOE 10.1: Vortrag
Mittwoch, 11. März 2026, 15:00–15:15, GÖR/0226
Combining machine-learning and dynamic network models for sepsis prediction — •Juri Backes1,3, Artyom Tsanda1,2, Tobias Knopp1,2, Wolfgang Renz3, and Eckehard Schöll4 — 1TU Hamburg — 2UKE Hamburg — 3HAW Hamburg — 4TU Berlin
We enhance short-term sepsis predictions by integrating machine learning techniques like Auto-Encoders and Gated-Recurrent-Units with a dynamical 2-layer network model of adaptive phase oscillators [1] representing the interaction between parenchymal cells (functional organ cells) and the immune system via cytokines. The model trajectories determined by machine learning are used for detection and prediction of critical infection states and mortality. The model-based predictions are compared with those of purely data-based approaches in terms of predictive power and interpretability. To this end we project real high-dimensional medical patient data into the low-dimensional parameter space of the model.
[1] R. Berner, J. Sawicki, M. Thiele, T. Löser, and E. Schöll: Critical parameters in dynamic network modeling of sepsis. Front. Netw. Physiol. 2, 904480 (2022).
Keywords: Network Physiology; 2-layer network of oscillators; synchronization; machine learning; sepsis