ArterialNet: Arterial Blood Pressure Reconstruction

Accurate and continuous monitoring of arterial blood pressure (ABP) is vital for clinical hemodynamic monitoring. However, current methods are either invasive, requiring insertion of catheters, or provide limited information, lacking comprehensive ABP waveforms. Cuffless wearable solutions, combined with deep learning, offer potential but face challenges in accurately reconstructing ABP waveforms and estimating systolic and diastolic blood pressure (SBP/DBP) due to individual variability. We propose a custom pre-trained backbone and a tailored optimization function to address these challenges. Our method demonstrates superior performance in ABP waveform reconstruction and accurate SBP/DBP estimations, while significantly reducing subject variance. To validate the effectiveness of our approach, we conducted comprehensive evaluations using both in-clinic data and a pioneering study involving remote health monitoring with cuffless data. Our results surpass previous efforts, demonstrating a root mean square error (RMSE) of 5.41 ± 1.35 mmHg and a minimum of 58% lower standard deviation (SD) across all measurements. These outcomes highlight the robustness and precision of our method in accurately estimating SBP/DBP and reconstructing ABP waveforms. Furthermore, we assessed the performance of our solution in non-clinical settings using the CTRAL BioZ dataset. The evaluation yielded an RMSE of 8.66 ± 1.13 mmHg for ABP, proving the potential of ABP reconstruction under remote health settings.

动脉血压（ABP）的准确持续监测对于临床血流动力学监测至关重要。然而，当前的方法要么是有创的，需要插入导管，要么提供的信息有限，缺乏完整的ABP波形。无袖带可穿戴解决方案结合深度学习具有潜力，但由于个体差异，在准确重建ABP波形以及估算收缩压和舒张压（SBP/DBP）方面面临挑战。我们提出了一种定制的预训练主干网络和一种量身定制的优化函数来应对这些挑战。我们的方法在ABP波形重建和准确的SBP/DBP估算方面表现出优越性能，同时显著降低了个体差异。为了验证我们方法的有效性，我们使用临床数据以及一项涉及无袖带数据远程健康监测的开创性研究进行了综合评估。我们的结果优于以往的研究，在所有测量中均方根误差（RMSE）为5.41±1.35 mmHg，标准差（SD）至少降低了58%。这些结果凸显了我们的方法在准确估算SBP/DBP和重建ABP波形方面的稳健性和精确性。此外，我们使用CTRAL BioZ数据集评估了我们的解决方案在非临床环境中的性能。评估得出ABP的RMSE为8.66±1.13 mmHg，证明了在远程健康环境下ABP重建的潜力。