Fig. 4: Demonstration of mechanical and biopotential artefacts mitigating by metahydrogel.
a, Distribution of SBP and DBP data collected over 2 min under dynamic conditions with and without the use of metahydrogel. Data were obtained from one independent participant. The boxes represent the interquartile range (25th–75th percentile), while the whiskers indicate the mean ± standard deviation (s.d.). b, Bland–Altman plot showing HR measurement agreement between the metahydrogel ECG electrode and piezoelectric artery sensor. c, Stability evaluation of collected BP signals by the recurrent analysis of collected ECG signals in time domain. d, Schematic of the autoencoder framework for denoising. The autoencoder model denoises the data by mapping measurement-corrupted data points back to the data manifold (black arrows). The bottom part presents typical AP and ECG signals before and after the denoising process. e, Calculated recognition accuracy of artery pulse and ECG signal versus SNR. The bar plots show SNRs of sensing signals with different denoising process. f, Frequency–power spectra of ECG signals collected with and without the use of metahydrogel after processing with the autoencoder-based denoising algorithm. g, Stability evaluation of BP signals collected over 2 min during a kinetic activity. Data were obtained from one independent participant. Box plots show the median (central line), interquartile range (25th–75th percentile, box) and whiskers extending to the minimum and maximum values. h, Accuracy comparison of measured DBP achieved with and without metahydrogel (red star and red dot; the arrow serves as a guide to the eye) and other approaches. The grey region shows that these data were collected under static state, and the background shading indicates the IEEE accuracy categories (grades A, B and C) for wearable BP monitoring.
