Neural Network-Driven Counter-Pulsation in Pulsatile Extracorporeal Membrane Oxygenator(ECMO): Enhancing Real-time Pulse Discrimination and Control Efficiency

Abstract

Implementing counter-pulsation (CP) control in pulsatile extracorporeal membrane oxygenator (p-ECMO) systems offers a refined approach to mitigate risks commonly associated with conventional ECMOs. To attain CP between the p-ECMO and heart, accurate detection of heartbeats within blood pressure (BP) waveform data becomes imperative, especially in situations where measuring electrocardiograms (ECGs) are difficult or impractical. In this study, a cumulative algorithm incorporating filter-type neural networks was developed to distinguish heartbeats from other pulse signals generated by the p-ECMO, reflections, or motion artifacts in the BP data. A control system was implemented using the cumulative algorithm that detects the heart rate (HR) and maintains a proper interval between the p-ECMO's pulses and heart beats, thereby achieving CP. To ensure precise circulatory support control, the p-ECMO setup was connected to a mock circulation system, with the human BP waveforms being replicated using a heart model. The algorithm could maintain CP perfectly when the HR remained constant; however, owing to a 0.48-s delay from the HR detection to CP control, the success rate of the CP control decreases when a sudden increase in the HR occurred. In fact, when the HR varied by ± 5 bpm every minute, the CP success rate dropped to 78.62%, however this was still higher compared to the 25.75% success rate achieved when no control was applied.