A novel optical fiber sensor using a mesh microbend optical fiber sensor to measure the perioperative heart rate (HR) and respiratory rate (RR) frequency signals was developed by our team. The feasibility of the sensor was evaluated in 10 infants in the perioperative period. We used traditional algorithms, such as Fast Fourier Transformation (FFT) and Wavelet Transformation (WT) to remove the noise and extract the features of the acquired HR and RR signals. However, the nonlinear fitting abilities of those traditional algorithms failed to completely remove the noise hence it was difficult to extract the features effectively. In this paper, we propose a deep learning model EMD-LSTM-CNN (ELC) to process both HR and RR based on Long Short Term Memory (LSTM), Convolutional Neural Network (CNN), and Empirical Modal Decomposition (EMD) methods. The trend term is extracted by EMD from HR and RR. The CNN and LSTM are applied to extract features and process them respectively. The experimental results show that the deep learning model has a better result compared with the traditional FFT and WT algorithms. The proposed model demonstrates compliance with the current standard physiological monitoring method in measuring non-stationary vibration signals such as HR and RR, which promises potential clinical applications in the future.

中文翻译：

---

基于ELC模型的围手术期婴儿心率和呼吸频率评估

我们的团队开发了一种使用网状微弯光纤传感器测量围手术期心率 (HR) 和呼吸率 (RR) 频率信号的新型光纤传感器。在围手术期对 10 名婴儿进行了传感器的可行性评估。我们使用传统算法，如快速傅里叶变换 (FFT) 和小波变换 (WT) 来去除噪声并提取采集到的 HR 和 RR 信号的特征。然而，这些传统算法的非线性拟合能力并不能完全去除噪声，因此难以有效地提取特征。在本文中，我们提出了一种深度学习模型 EMD-LSTM-CNN (ELC)，基于长短期记忆 (LSTM)、卷积神经网络 (CNN) 和经验模态分解 (EMD) 方法来处理 HR 和 RR。EMD 从 HR 和 RR 中提取趋势项。CNN 和 LSTM 分别用于提取特征和处理它们。实验结果表明，与传统的FFT和WT算法相比，深度学习模型具有更好的效果。所提出的模型证明了在测量 HR 和 RR 等非平稳振动信号时符合当前的标准生理监测方法，这有望在未来的临床应用中得到应用。