Continuous non-invasive arterial blood pressure monitoring with photoplethysmography via SE-MSResUNet network

Abstract

ObjectiveArterial blood pressure (ABP) waveforms include physiological parameters closely related to BP values, while only a few studies predict BP values through the monitoring ABP waveforms. Current ABP waveforms monitoring methods are limited by its high computing cost and information redundancy of multi-channel input signals. The aim of our work is to propose a new end-to-end method for monitoring the ABP waveforms using single-channel photoplethysmography (PPG) signals by constructing a lightweight Multi-Scale Residual U-Net fused with Squeeze-and-Excitation module (SE-MSResUNet) model. MethodsFirstly, the PPG and ABP signals are preprocessed with denoising, segmentation, and eliminating phase difference. Subsequently, the SE-MSResUNet model is constructed and trained using training data to achieve the mapping relationship between PPG and ABP signals. Then, the BP values are extracted from the monitored ABP waveforms. Finally, the performance of the SE-MSResUNet in ABP waveform monitoring and BP estimation is investigated. ResultsThe ABP waveforms monitored by our proposed SE-MSResUNet network with PPG signals only are highly consistent with the reference ABP waveforms. The MAE ± SD of estimated diastolic blood pressure (DBP), mean arterial pressure (MAP) and systolic blood pressure (SBP) compared with reference values are 2.16 ± 3.75, 2.29 ± 3.72, and 3.88 ± 6.17 mmHg, respectively. The proposed method meets the Association for the Advancement of Medical Instrumentation (AAMI) standard and reach A level of the British Hypertension Society (BHS) standard. ConclusionsOur proposed method outperforms current state-of-the-art methods for non-invasive blood pressure monitoring from PPG signals. The algorithm provides a potential possibility for continuous non-invasive blood pressure monitoring through portable wearable devices.