Venous Pressure Prediction for ECMO Based on GWO-VMD-LSTM-Informer

Extracorporeal membrane oxygenation (ECMO) is a critical technology for treating patients with acute heart and respiratory failure; however, the venous pressure signal fluctuates significantly during treatment, posing challenges for monitoring and prediction. This study proposes a venous pressure prediction model based on GWO-VMD-LSTM-Informer to achieve high prediction accuracy. First, the Random Forest algorithm is applied to select features highly correlated with venous pressure. Then, the Variational Mode Decomposition (VMD) parameters are optimized using the Grey Wolf Optimizer (GWO) to extract the optimal components of the pressure signal. The study then incorporates Sample Entropy (SE) to differentiate between low and high entropy signals and employs Informer and Long Short-Term Memory (LSTM) models to predict signals corresponding to the various components. Finally, the prediction results for each component are synthesized to generate the overall venous pressure prediction. Experimental results obtained from sheep in the low-flow group under the veno-venous extracorporeal membrane oxygenation (VV-ECMO) model demonstrated that the proposed model significantly outperformed other models, achieving low absolute error (MAE = 1.011), root mean square error (RMSE = 1.420), and a high coefficient of determination (R2= 0.97) in predicting venous pressure for the next 5 minutes. These results suggest that the GWO-VMD-LSTM-Informer model can effectively predict non-stationary signals, providing valuable support for clinical decision-making and improving treatment accuracy.