Wearable Hyperspectral Photoplethysmography Allows Continuous Monitoring of Exercise‐Induced Hypertension

Figures

• 

  Figure 1. A schematic illustration for continuous monitoring of blood pressure using wearable hyperspectral photoplethysmography (PPG) sensor with double‐folded solid immersion grating microspectrometer (DFSIG‐µSPEC). a) The wearable hyperspectral PPG module comprises DFSIG‐µSPEC and a white LED. Light emitted from the LED penetrates the skin and undergoes reflection. The reflected light passes through a silicon microslit, dispersed through the solid immersion grating (SIG), undergoing multiple reflections and refractions with plane mirrors and a convex lens. Dispersed light delivers PPG signals on a CMOS line sensor at the level of hyperspectral range. Measured spectra are accumulated into 3D PPG datasets over time. Hyperspectral PPG signals allow the precise calculation of continuous arteriolar pulse transit time (aPTT) for blood pressure estimation. b) The microfabrication steps and module integration of DFSIG‐µSPEC. The SIG is microfabricated by using wafer stepper lithography and reactive ion etching (step i), and metallized with an aluminum thin layer (step ii). A convex lens is molded with PDMS (step iii). The SIG and a CMOS line sensor are mounted on a printed circuit board (PCB),…

• 

  Figure 2. Spectral resolution of DFSIG‐µSPEC. a) Output spectra of Nd: YAG tunable laser measured by DFSIG‐µSPEC. Each spectrum exhibits a sharp spectral peak in the operational range of DFSIG‐µSPEC. b) Spectral resolution and optical sensitivity. DFSIG‐µSPEC exhibits an average resolution of 3.4 nm and a relative sensitivity of over 60% of the maximum. c) Spectrum measurement of white LED light, where both the commercial spectrometer and DFSIG‐µSPEC are nearly identical. d) Output spectra of narrow band LEDs with spectral peaks at 588, 633, 675, and 785 nm. Measured spectra are well‐matched with those from the commercial spectrometer. e) Emission spectra of 1 µM‐concentration of ROX, Cy5, and mixed solutions. Two spectral peaks exhibit well‐defined peaks with a 78% overlap intensity. f) Reflectance spectra at the wrist skin. The measured spectrum of DFSIG‐µSPEC exhibits similar spectral profiles of the reference spectrum.

• 

  Figure 3. Hyperspectral PPG (HS‐PPG) waveform measurement for continuous blood pressure estimation. a) Blood pressure estimation flowchart. Light is passed through the vascular layers of the epidermis (EPI), dermis (DER), and subcutaneous tissue (SUB) and absorbed in the blood vessels. Systemic vascular resistance (SVR), i.e., a measure of resistance to blood flow, differs from the vascular layers. A vascular model is simplified with different SVR values. Systolic/diastolic blood pressures are finally estimated from the SVR values after calculation of the mean blood pressure (MBP) and pulse pressure (PP). b) Time‐lapsed HS‐PPG waveforms (sampling rate: 1 kHz). Continuous aPTT values are calculated using the time difference for the shortest wavelength of 550 nm. c) The time delay between the PPG peaks using the 550 nm peak as the reference and some examples of PPG waveforms. The PPG waveform exhibits higher time delay as the wavelength increases, due to deeper penetration depth. d) A simplified vascular model for continuous aPTT. The vascular layers are replaced by an equivalent electrical circuit with SVR and arterial compliance (AC), which are affected by the wavelength. The simplified…

• 

  Figure 4. Comparative analysis of blood pressure measurement in normotensive states. a) A strong linear correlation of DBP values from cuff‐based BP sensor and HS‐PPG module. The estimated BP is monitored through HS‐PPG module, whereas the reference BP is measured by a cuff‐based BP sensor. DBPs show high linearity with an R‐value of 0.92 and the mean absolute difference (MAD) of 1.20 mmHg. (n = 60, p < 0.05) b) The Bland–Altman plot for DBP values. c) A strong linear correlation of SBP values from cuff‐based BP sensor and HS‐PPG module. SBPs also show high linearity with an R‐value of 0.96 and the MAD of 0.40 mmHg. (n = 60, p < 0.05) d) The Bland–Altman plot for SBP values. The mean estimation error is ‐0.07 mmHg for DBP and 0.04 mmHg for SBP, with confidence intervals over 93% for each. e) The R‐value and MAD in HS‐PPG waveform depending on the number of spectral bands (the operational range over the wavelength interval), limited by the spectral resolution of DFSIG‐µSPEC. High spectral resolution significantly enhances estimation reliability, increasing the R‐value of SBP from 0.81 to 0.94 and DBP from 0.78 to 0.92. MAD also reduces from 9.04 to 2.94 mmHg of SBP and 9.50–3.62 mmHg of…

• 

  Figure 5. Continuous BP monitoring during hemodynamic changes. a) Experimental setup for continuous BP monitoring. HS‐PPG waveforms, BP, heart rate (HR), and respiratory exchange ratio (RER) were simultaneously monitored with respective sensors. b) A representative example of SBP, HR, and RER signals, recorded from a single subject. All the experiments are performed with three steps (10‐min rest, 10‐min exercise, and 10‐min recovery). Both reference and estimated SBPs significantly increase during exercise and decreased during recovery. SBP is measured at 1‐min interval during exercise with the HS‐PPG module, unlike cuff‐based BP sensor. Note that cuff‐based BP is inaccurate during exercise due to motion artifact. HR sharply increases during exercise and decreases during recovery, similar to SBP. RER slightly decreases immediately after exercise and then overshoots during recovery. HS‐PPG module clearly exhibits immediate responses of SBP during hemodynamic changes like the performance of the HR sensor or calorimeter. c) A strong linear correlation between reference and estimated SBPs during the recovery step. Estimated SBP exhibits a high correlation (R2 = 0.95) with the reference SBP (n…