Abstract
Extracorporeal oxygenation supports patients with severe cardiac or respiratory failure, with the oxygenator providing critical gas exchange. Thrombus formation in the oxygenator can impair efficiency and increase risks such as hemolysis and embolism, but existing detection methods are limited in accuracy and timeliness. This study introduces a computational bioimpedance approach for early thrombus detection that integrates advanced modeling and machine learning techniques while preserving the oxygenator's functionality. We developed a finite element model of an oxygenator to simulate bioimpedance measurements using varied electrode configurations. Neural networks optimized electrode placement and injection-measurement patterns, enhancing sensitivity to conductivity changes. A second neural network was trained on simulated data to distinguish between normal and thrombus-affected conditions, achieving an F1-score exceeding 94% in classification tasks. Simulations demonstrated the feasibility of this method, with optimized configurations significantly improving detection accuracy. The findings suggest that computational bioimpedance, combined with neural network optimization, provides a robust framework for automated thrombus detection inside an oxygenator.