A new portable blood flow volume (BFV) sensor is developed from the function of estimating oxygen saturation (SpO₂) implemented inherently to maximize the accuracy of predicting BFV. The sensor is designed to estimate BFV in high accuracy at the arteriovenous fistula (AVF) of a hemodialysis (HD) patient based on a built artificial neural network (ANN). The BFVs measured by the proposed sensor would help greatly evaluate the AVF complications at early stage, such as infection, bleeding, stenosis, and vascular calcification, while AVFs are under long-time usage. The sensor module consists of LEDs/PDs, readout circuitry and algorithm to estimate BFV. The oxygen saturation (SpO₂) is also estimated using the same hardware to serve as one of input features for the aforementioned ANN for maximize the accuracy in BFV. With using the hardware, oxygen saturation’s algorithm is built and implemented with the mean deviation (MD) and standard deviation (SD) adapted FDA standard, which is MD ∓ 1.96SD = 0.024% ∓ 1.772% < 5%. Besides, the BFV ground-truth data are obtained from a transonic HD03 dilution machine for calibrating the ANN. The accuracy in estimating BFV with and without estimating SpO₂ are compared, which correspond to R² = 0. 94,517, root mean square error = 92.199 ml/min and R² = 0.87145, root mean square error = 141.015 ml/min, respectively. Besides, another neural network model is implemented with fixing value of an input feature of SpO₂, the ANN model’s correlation reaches only 0.1554, while the related root mean square error is 493.2442 ml/min. Therefore, the results validate clearly the necessity of the inherent estimation on SpO₂.

通过固有地估算氧饱和度（SpO ₂）的功能，开发了一种新的便携式血流量（BFV）传感器，以最大化预测BFV的准确性。该传感器旨在基于内置的人工神经网络（ANN），以高准确度估计血液透析（HD）患者动静脉瘘（AVF）处的BFV。建议的传感器测量的BFV有助于在早期评估AVF并发症，例如感染，出血，狭窄和血管钙化，而AVF则需要长期使用。传感器模块由LED / PD，读取电路和估算BFV的算法组成。氧饱和度（SpO ₂还使用相同的硬件来估算）作为上述ANN的输入功能之一，以最大程度地提高BFV的精度。通过使用硬件，可以建立和实施氧饱和度算法，并采用符合FDA标准的平均偏差（MD）和标准偏差（SD），即MD∓1.96SD = 0.024％∓1.772％<5％。此外，BFV地面真相数据是从跨音速HD03稀释机获得的，用于校准ANN。比较了在估计和未估计SpO _2的情况下BFV的估计准确度，它们对应于R ²  =0。94,517，均方根误差= 92.199 ml / min和R ² = 0.87145，均方根误差= 141.015 ml / min。此外，利用SpO ₂输入特征的固定值实现了另一个神经网络模型，ANN模型的相关性仅为0.1554，而相关的均方根误差为493.2442 ml / min。因此，结果清楚地证实了对SpO ₂进行固有估计的必要性。