A noninvasive time-frequency-based approach to estimate cuffless arterial blood pressure

Abstract

Arterial blood pressure (ABP) is one of the most vital signs in the prophylaxis and treatment of blood pressure-related diseases because raised blood pressure is the most significant cause of death and the second major cause of disability in the world. Higher ABP yields greater strain on arteries and these extra strains turn arteries into thicker, less flexible, and more narrow structures. This increases the possibility of having an artery busting or artery occlusion, which are the primary reasons for heart attacks, kidney disease, or strokes. In addition to its importance in monitoring cardiovascular homeostasis, measurement of ABP is imperative in surgical operations. In this study, a simple and effective approach was proposed to estimate ABP from electrocardiogram (ECG) and photoplethysmograph (PPG) signals by an extreme learning machine (ELM) and statistical properties of the ECG and/or PPG signals in the time-frequency domain. To evaluate and apply the proposed approach, the Cuffless Blood Pressure Estimation Dataset, which was published and shared by UCI, was employed. First, the statistical properties were extracted from ECG and PPG signals that were in the time-frequency domain. Later, extracted features were employed to estimate cuffless ABP for each subject by the ELM and some popular machine learning methods. Achieved results and reported results in the literature showed that the proposed approach can be successfully employed for estimating cuffless blood pressure (BP) from ECGs and/or PPGs. Additionally, with the proposed approach, the systolic BP, mean BP, and diastolic BP can be calculated simultaneously.

End-to-End Blood Pressure Prediction via Fully Convolutional Networks

“…Wang [10] segments a single PPG signal from the raw PPG signal, extracts morphological and spectral features from the signal, and performs experiments to predict SBP and DBP through artificial neural networks (ANN), which achieved SBP 4.02 ± 2.79 and DBP 2.27 ± 1.82. Ertugrul [11] uses a spectrogram, which is the magnitude squared of the short-time Fourier transform (STFT) of a PPG and/or ECG signal, and then performed a study to predict BP through the extreme learning machine method (ELM). The accuracy of BP prediction was found to be SBP 4.37 and DBP 3.95.…”

“…More precisely, since the inputs to our model have sizes of 512 and 256 for the time and frequency domain respectively, the kernel sizes of each Extraction (k EXT ) and Concentration blocks (k CON ) are needed to satisfy these constraints. Possible combinations of the filter sizes for our 4-EC stacked networks are (3,27,25), (5,19,17) and (7,11,9) for (k EXT , k t CON , k f CON ), where k t CON and k f CON are the filter sizes of the Concentration block for time and frequency flow. We selected (7,11,9) for the experiments, and related results regarding filter sizes will be described in Section IV-B.…”

“…Possible combinations of the filter sizes for our 4-EC stacked networks are (3,27,25), (5,19,17) and (7,11,9) for (k EXT , k t CON , k f CON ), where k t CON and k f CON are the filter sizes of the Concentration block for time and frequency flow. We selected (7,11,9) for the experiments, and related results regarding filter sizes will be described in Section IV-B.…”

“…Training without consideration of train/test data split by subject would result in an abnormally high BP prediction accuracy because the subject data used in the training phase may be used in the test phase. For a fair comparison, we did not compare the performance with the results of studies not considering train/test data split by subject [4], [9], [10], [11] and the study using private dataset [4]. We compared only the results obtained under the same conditions in this study.…”

“…As another way to predict BP without a cuff, some researchers have attempted to predict BP using a single raw biomedical signal or transform rather than the PWV-based feature [9], [10], [11], [12]. The concept of predicting BP using a single signal measurement such as ECG or PPG appears to be more appropriate for mobile devices such as smartphones and smart watches.…”