Aortic Pressure Waveform Estimation Based on Variational Mode Decomposition and Gated Recurrent Unit

Objective: Aortic pressure waveform (APW) can provide vital indices for the diagnosis of cardiovascular diseases. Although various APW estimation methods have been reported to avoid the risks of direct invasive measurement, more accurate and practical estimation models still need to be developed to promote the application of APW in routine monitoring. To solve this problem, a hybrid model based on variational mode decomposition (VMD) and gated recurrent unit (GRU) named VMD-GRU was proposed to estimate the APW from the brachial pressure waveform (BPW). Methods: Invasive APWs and BPWs from 34 subjects were used to validate the proposed hybrid model. Initially, new samples were obtained from these measured APWs and BPWs using data augmentation technology. Subsequently, VMD was employed to decompose the augmented BPWs into multiple intrinsic mode functions (IMFs). Next, the GRU network was utilized to learn the complex relationship between the IMFs and augmented APWs. The trained GRU network can be used to estimate the APWs with the IMFs obtained from the BPWs and VMD approach. The root-mean-square-error of total waveform and mean absolute errors of commonly adopted hemodynamic indices in clinic including systolic and diastolic blood pressures and pulse pressure were used to evaluate the proposed hybrid model. The performance of the proposed hybrid model was obtained from leave-one-subject-out cross validation. Results: The proposed hybrid model achieved smaller errors of total waveform (3.33 vs. 3.50 mmHg, P < 0.05) and pulse pressure (3.02 vs. 3.64 mmHg, 0.05 < P < 0.10) compared to the GRU. Conclusion: The proposed hybrid model can provide more accurate APW in comparison to the GRU network.