Camera-based assessment of hemodynamics for non-contact blood pressure estimation

该项目旨在利用相机进行非接触式血压测量。为此，我们开发了一种普遍适用的方法来评估视频中的血流动力学功能，并在系统实验中分别评估了所提出的方法及其用于血压估计的可行性。最近的研究证明(1)心脏活动的多模态，即通过血容量脉冲和弹道心动图效应，影响视频记录；(2)视频记录允许对血液脉动进行时空评估。我们利用这两个方面来获得与近端（通过弹道心动图）和远端（通过脉冲波到达和脉冲波传播）心血管活动相关的参数，并从这些参数中估计搏动血压。在该项目中，我们首先开发了从视频中提取形状保留的方法，即形态学可用的光容积脉搏图（cbPPG）和弹道心动图（cbBCG）。所提出的方法利用非线性算法的时空平均进行信号扭曲（二维信号扭曲）。基于空间分辨cbPPG，我们开发了一种利用视频（所谓的灌注速度）估计周围血管张力的新方法。灌注速度通过概率方法获得，该方法确定外周脉冲波速度的空间分解概率分布，并从局部分布的叠加中估计灌注速度。为了最终估计血压，我们将脉搏传递时间（由cbBCG和cbPPG得出）与多水平分析中的部分互补灌注速度结合起来，确定了非接触血压估计模型。为了验证所提出的方法，我们进行了一项实验研究，包括50名健康受试者，他们接受了为该项目制定的实验方案。因此，该项目提供了使用相机对血流动力学特性进行非接触评估的方法。这些方法可以用于估计搏动间血压，但即使超出血压估计，它们也为评估灌注和自主调节提供了广泛的机会。我们的基础研究进一步为描述基于相机的关键参数评估的机会和局限性做出了重大贡献，特别是与脉冲波传播的时空特征有关。

The project aims at the non-contact blood pressure estimation by using cameras. To that end, we develop generally applicable methods to assess the hemodynamic function from videos and evaluate the proposed methods and their feasibility for blood pressure estimation, respectively, within systematic experiments.Recent works proved (1) that the heart activity multimodally, i.e. by the blood volume pulse and ballistocardiographic effects, affects video recordings and (2) that video recordings allow the spatio-temporal assessment of blood pulsations. We exploit both aspects to derive parameters related to the proximal (via ballistocardiography) and distal (via pulse wave arrival and pulse wave propagation) cardiovascular activity and estimate the beat-to-beat blood pressure from such parameters. Within the project, we firstly develop methods to extract shape-preserved, i.e. morphologically usable, photoplethysmograms (cbPPG) and ballistocardiograms (cbBCG) from videos. The proposed methods make use of spatio-temporal averaging using a non-linear algorithm for signal warping (two-dimensional signal warping). Based on spatially resolved cbPPG, we develop a novel method to estimate the peripheral vascular tone using videos (the so-called perfusion velocity). The perfusion velocity is gained by a probabilistic approach, which determines spatially resolved probability distributions of the peripheral pulse wave velocity and estimates the perfusion velocity from the superposition of local distributions. To eventually estimate the blood pressure, we combine the pulse transit time, which is derived from cbBCG and cbPPG, and the partially complementary perfusion velocity within a multilevel analysis and determine a model for non-contact blood pressure estimation. To validate the proposed methods, we conduct an experimental study comprising 50 healthy subjects undergoing an experimental protocol that was developed for the project. As a result, the project provides methods for the non-contact assessment of hemodynamic characteristics using cameras. Such methods can be applied to estimate the beat-to-beat blood pressure, but even beyond blood pressure estimation, they offer wide opportunities to assess the perfusion and autonomous regulation. Our basic investigations further make a major contribution to characterize the opportunities and limitations of camera-based assessment of vital parameters, particularly related to the spatio-temporal characteristics of pulse wave propagation.