Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 7: Modeling and Data Analysis
DY 7.5: Vortrag
Montag, 12. März 2018, 11:00–11:15, BH-N 128
Breathing with the beating of the heart: A machine-learner's approach to ECG-derived respiratory signal estimation — •Stephan Bialonski1, Daniel Vorberg2, and Justus Schwabedal3 — 1Center for Advancing Electronics Dresden (cfaed), TU Dresden — 2Max-Planck-Institute for the Physics of Complex Systems, Dresden — 3Department of Biomedical Informatics, Emory University School of Medicine, Atlanta
We investigated how machine learning models can be used to extract knowledge from biophysical systems. As a test case, we studied the challenge to derive respiratory information from electrocardiographic (ECG) signals, a long-standing problem in sleep research. We identified two well-known coupling mechanisms by analyzing a long short-term memory (LSTM) architecture that we fitted to predict respiratory information from ECG signals. These mechanisms couple heart beat dynamics and respiration and comprise of a physiological coupling (respiratory sinus arrhythmia) as well as a physical coupling that is related to the position of measurement electrodes relative to the heart. We verified these results by modelling the coupling mechanisms and studying the resulting patterns in our LSTM architecture.