Abstract Today’s methods for estimating respiratory rate (RR) from the electrocardiograph (ECG) and the photoplethysmogram (PPG) are not good at distinguishing between periods of low- and high-quality data or raw signals. The goal of this work is to present an alternative way of estimating respiratory rate from ECG and PPG by using machine learning to improve the accuracy of estimation. The datasets used in this work are extracted from a publicly available source, the BIDMC dataset. The proposed methods are based on respiratory signals extracted from raw signals and use support vector machine (SVM) and neural networks (NNs) to estimate respiratory rate. The performance using a window size (32 s) is compared with that of current methods. The proposed methods achieved comparable accuracy to current methods when the number of classes is low. Once the number of classes increases, the accuracy drops significantly. This work demonstrates that the use of machine learning has more potential than commonly thought, and it is necessary to apply it and carry out further studies and research on existing methods.
Paper
Full text
Estimating the respiratory rate from ECG and PPG using machine learning techniques
Semantic Scholar · Medicine · 2021
Abstract
Abstract Today’s methods for estimating respiratory rate (RR) from the electrocardiograph (ECG) and the photoplethysmogram (PPG) are not good at distinguishing between periods of low- and high-quality data or raw signals. The goal of this work is to present an alternative way of estimating respiratory rate from ECG and PPG by using machine learning to improve the accuracy of estimation. The datasets used in this work are extracted from a publicly available source, the BIDMC dataset. The proposed methods are based on respiratory signals extracted from raw signals and use support vector machine (SVM) and neural networks (NNs) to estimate respiratory rate. The performance using a window size (32 s) is compared with that of current methods. The proposed methods achieved comparable accuracy to current methods when the number of classes is low. Once the number of classes increases, the accuracy drops significantly. This work demonstrates that the use of machine learning has more potential than commonly thought, and it is necessary to apply it and carry out further studies and research on existing methods.