BACKGROUND Cuff-based blood pressure measurement lacks comfort and convenience. Here, we examined whether blood pressure can be determined in a contactless manner using a novel smartphone-based technology called transdermal optical imaging. This technology processes imperceptible facial blood flow changes from videos captured with a smartphone camera and uses advanced machine learning to determine blood pressure from the captured signal. METHODS We enrolled 1328 normotensive adults in our study. We used an advanced machine learning algorithm to create computational models that predict reference systolic, diastolic, and pulse pressure from facial blood flow data. We used 70% of our data set to train these models and 15% of our data set to test them. The remaining 15% of the sample was used to validate model performance. RESULTS We found that our models predicted blood pressure with a measurement bias±SD of 0.39±7.30 mm Hg for systolic pressure, -0.20±6.00 mm Hg for diastolic pressure, and 0.52±6.42 mm Hg for pulse pressure, respectively. CONCLUSIONS Our results in normotensive adults fall within 5±8 mm Hg of reference measurements. Future work will determine whether these models meet the clinically accepted accuracy threshold of 5±8 mm Hg when tested on a full range of blood pressures according to international accuracy standards.

中文翻译：

---

使用透皮光学成像技术的基于智能手机的血压测量。

背景技术基于袖带的血压测量缺乏舒适性和便利性。在这里，我们检查了是否可以使用一种称为透皮光学成像的基于智能手机的新型技术以非接触方式确定血压。这项技术可以处理智能手机相机拍摄的视频中无法察觉的面部血流变化，并使用高级机器学习来根据拍摄的信号确定血压。方法我们在研究中招募了1328名血压正常的成年人。我们使用了一种先进的机器学习算法来创建计算模型，该模型可以根据面部血流数据预测参考收缩压，舒张压和脉搏压。我们使用了70％的数据集来训练这些模型，并使用了15％的数据集来对其进行测试。其余15％的样本用于验证模型性能。结果我们发现我们的模型预测血压的测量偏差±SD的收缩压为0.39±7.30 mm Hg，舒张压为-0.20±6.00 mm Hg，脉压为0.52±6.42 mm Hg。结论我们在正常血压成年人中的结果在参考测量值的5±8 mm Hg之内。未来的工作将确定这些模型在根据国际精度标准在全范围的血压上进行测试时是否满足临床认可的5±8 mm Hg的精度阈值。