Non-invasive estimation of beat-by-beat aortic blood pressures from electrical impedance tomography data processed by machine learning

• DOI: 10.1007/s10877-025-01274-2
• Publication Date: 2025-03-27T23:28:41Z
• Authors: Fabian Müller-Graf; Jacob P. Thönes; Lisa Krukewitt; Paul Frenkel; Henryk Richter; Sascha Spors; Volker Kühn; Amelie R. Zitzmann; Stephan H. Boehm; Daniel A. Reuter

Abstract Hypotension in perioperative and intensive care settings is a significant risk factor associated with complications such as myocardial infarction and kidney injury thereby increasing perioperative complications and mortality. Continuous blood pressure monitoring is essential, yet challenging due to the invasive nature of current methods. Non-invasive techniques like Electrical Impedance Tomography (EIT) have been explored but face challenges in accurate and consistent blood pressure estimation. A machine learning (ML) approach was used to predict aortic blood pressures from EIT voltage measurements in landrace pigs. A convolutional neural network (CNN) was trained on a dataset of 75 298 heartbeats, to predict systolic (SAP), mean (MAP), and diastolic arterial pressures (DAP) of individuals whose arterial pressures were unknown to the algorithm. The Intraclass Correlation Coefficient (3,1) with absolute agreement (ICC) was calculated and the concordance was estimated, comparing reference blood pressure measurements and ML-derived estimates. A risk classification was estimated for the calculated blood pressure as suggested by Saugel et al. 2018. The ML-model demonstrated moderate correlations with invasive blood pressure measurements (ICC for SAP of 0.530, for MAP of 0.563, and for DAP of 0.521.) with a low risk score for 75.8% of the SAP and 64.2% of MAP estimated blood pressures. ML-techniques using EIT-voltages showed promising preliminary results in non-invasive aortic blood pressure estimation. Despite limitations in the amount of available training data and the experimental setup, this study illustrates the potential of integrating ML in EIT signal processing for real-time, non-invasive blood pressure monitoring.