Design, fabrication, and biomechanical compatibility of the multimodal sensor
As illustrated in Fig. 1a, the multimodal photonic sensor enables simultaneous on-skin monitoring of various physiological signals associated with strain, temperature, and pH, providing comprehensive and personalized healthcare information for improved disease diagnosis and treatment. The multimodal sensor primarily consisted of a core/coating HPOF that was encapsulated between two elastic polyurethane layers of a skin-adhesive base and a water-proof protective cover, together with cotton fabric tape to secure the ends of the HPOF. The skin-adhesive base layer was designed with a narrow slit to enable direct contact of the HPOF with skin for sweat monitoring. Silica multimode optical fibers (MMFs) were pigtailed to the HPOF with their central axis aligned for laser coupling and emission collection. The core of the HPOF was made from composites of Ln-UCNPs and PDMS elastomers, and fabricated by approaches of molding and thermally curing (see Figure S1a, and more details in Materials and Method). Transparent, low-RI polyacrylamide (PAM) hydrogels were chosen as the coating of the HPOF, which endowed the fiber with step-index light guidance and meanwhile introduced abundant active amino groups for functional modifications. Since PDMS is hydrophobic, and the PAM hydrogels are hydrophilic polymers inflated with water, it is hardly to create strong adhesion between the PDMS and PAM via direct physical coating. To achieve a robust hydrogel coating, we chemically anchored the PAM hydrogels on PDMS by using an interfacial interpenetration strategy [25]. Briefly, the PDMS core was first treated with an ethanol-based organic solution containing hydrophobic photoinitiators (benzophenone) that would swell the PDMS, impelling diffusion and absorption of the photoinitiators into the core surface (Figure S1b). Afterwards, the treated core was immersed in a PAM precursor that contains hydrophilic photoinitiators (Irgacure 2959). Upon ultraviolet (UV) exposure, the hydrophilic photoinitiators induced polymerization of the PAM monomers while the hydrophobic initiators enabled the PAM networks to covalently crosslink with the PDMS chains, resulting in the formation of a thin hydrogel layer toughly bonded on the PDMS surface. It was found that the hydrogel coating could endure vigorous deformation induced by stretching or scratching with no observable crack or delamination, demonstrating excellent mechanical robustness (Figure S2). The length of the HPOF was optimized at ~ 1.5 cm from the profile of emission decay along the fiber length (Figure S3).
The multimodal photonic sensor integrates three distinct sensing mechanisms in a single HPOF to achieve simultaneous detection of strain, temperature, and pH (Fig. 1b, c). Upon near-infrared (NIR) excitation, the incorporated Ln-UCNPs in the HPOF generate multiband visible emissions via the upconversion process for versatile interrogation of different sensing mechanisms. The UCNPs were comprised of a luminescent core of NaYF4:Yb,Er covered by an inert shell of NaYF4, which could protect the core from luminescence quenching caused by surface defects and surrounding environments (Figure S4a) [26]. The morphology and particle size of the UCNPs were examined by transmission electron microscopy (TEM) and scanning electron microscope (SEM), where the UCNPs were uniformly distributed with hexagonal shape and diameter of ~ 57 nm (Fig. 1d, and Figure S4b-d). The Er3+ ions of the UCNPs possessed a pair of thermally coupled energy levels (2H11/2 and 4S3/2) [27], and the emission bands originating from these levels showed thermal-dependent behaviors, enabling sensitive temperature measurement from changes of the emission intensities. The PAM coating of the HPOF offers abundant amino groups allow facile chemical modifications with various functional molecules. Hereby, we functionalized the hydrogel coating with pH sensitive fluorescent dyes (pHrodo Red) to achieve pH responsiveness. The pHrodo Red was selected because of the substantial overlap between its absorption spectrum and the emission bands of the UCNPs, thereby enabling radiative energy transfer (RET) from the UCNPs to the pHrodo Red for pH determination (Figure S4e). Besides pH sensitivity, the loaded dye molecules also endowed the HPOF with wavelength-dependent absorption characteristics that could be further harnessed to quantify strain stimuli from the absorption changes of light propagating along the HPOF. Figure 1e shows a cross-section image of the HPOF (core/coating, 500 µm/553 µm) with red dyes loaded in the hydrogel coating. To avoid leakage of the dye molecules from the coating, the pHrodo Red modified with succinimidyl (NHS) ester was immobilized by covalent bonding to the amino groups of the hydrogel matrix (Figure S5).
The sensing HPOF was ultimately encapsulated in an elastic skin-adhesive patch to ensure stable adhesion on skin. Figures 1f and g show the images of the HPOF-based multimodal sensor conformally attached to the human skin, where the sensor could be freely stretched, compressed, and twisted without any failure. Extensive mechanical test was further conducted that confirmed the skin-like softness (Young’s modulus of 1.14 Mpa) and high stretchability (> 60%) of the sensor (Figure S6). Biological safety is critical for wearable sensors especially when in direct contact with human skins. The multimodal sensor was proven to be nontoxic and biocompatible as verified by a cell viability test with human neuroblastoma cells (SK-N-SH), where a high cell viability of > 98% was observed after 72 h of culture in the presence of the sensor (Figure S7). Figure 1h shows a typical emission spectrum of the sensor, indicating four separate emission bands generated upon excitation of a single 980-nm laser. The emission at 600 nm was produced by RET from the UCNPs to the pHrodo Red, while the emissions centered at 525, 545, and 655 nm were attributed to the Er3+ transitions of 2H11/2→4I15/2, 4S3/2→4I15/2, and 4F9/2→4I15/2 levels, respectively (Fig. 2a). With the distinctive design of materials, structures, and functionalities above, the multimodal sensor holds great promises as a wearable safety device for multi-stimuli sensing in long-term care settings.
Temperature sensing performances of the multimodal sensor
Temperature is one of the fundamental parameters in monitoring human activities and assessing body abnormalities [28]. To evaluate the feasibility of the multimodal sensor for temperature sensing, we characterized the temperature response of the sensor with a simplified optical setup as depicted in Figure S8. A fiber-coupled laser at a wavelength of 980 nm and an output power of 15 mW was launched into the sensor through the pigtailed silica MMFs, and the emissions were collected and guided to a compact spectrometer for spectral analysis. Figure 2b shows the emission spectra of the sensor in response to different temperatures. Because of the thermal coupling of 2H11/2 and 4S3/2 levels in Er3+, the emissions at 525 nm and 545 nm were strongly temperature-dependent and the intensity ratio of these two emissions followed the well-known Boltzmann distribution given by [27]:
$${I_{525}}/{I_{545}}=A\exp ( - \Delta E/kT)$$
1
Where \({I_{525}}\)and \({I_{545}}\) are emission intensities arising from the transitions of 2H11/2→4I15/2 and 4S3/2→4I15/2, respectively; is a constant; \(\Delta E\)denotes the energy gap between the 2H11/2 and 4S3/2;is the Boltzmann’s constant, and is the absolute temperature in the Kelvin scale. Figure 2c shows a linear plot of \(\ln ({I_{525}}/{I_{545}})\) versus the inverse temperature (\(1/T\)) over the temperature range of 25 ℃ (3.35×10− 3 K− 1) to 45 ℃ (3.14×10− 3 K− 1), which agrees well with Eq. (1). The temperature sensitivity defined as the percent change in \(\ln ({I_{525}}/{I_{545}})\) per unit change in temperature, was calculated to be 0.8% ℃−1 around 37 ℃ from the response curve, and the limit of detection (LOD) was about \(\pm\)0.19 ℃, estimated from the noise standard deviation of the temperature readout (Figure S9a). The response speed of the sensor was then studied upon a step change in temperature, where the response and recovery times of the sensor were measured to 3 s and 4 s, respectively (Fig. 2d). The highly sensitive and fast responding merits enabled the sensor to capture rapid subtle thermal signals produced by human activities such as nose breathing (Figure S9b). To evaluate the repeatability of the sensor, consecutive thermal cycling tests of heating and cooling were performed (Fig. 2e). The sensor showed reversable and reproducible readout with the cyclic temperature changes, indicating stable and repeatable performances.
We further validated the capability of the sensor to quantitatively detect dynamic thermal signals by a real-time temperature monitoring test, where the sensor was treated with water droplet of various temperatures (Fig. 2f). Figure 2g shows the temporal evolution of the droplet temperature measured by our sensor and a commercial IR camera. Notably, there was a high consistency between the temperature readout of our sensor and the signal obtained by the IR camera with a root mean square error (RMSE) of 0.46 ℃, demonstrating high reliability and accuracy in real-time temperature measurements. In addition, the influences of strain and pH stimuli on the temperature response of the sensor were also examined (Fig. 2h). The sensor was treated with different stretching strains and pH, separately, while the applied temperature was kept constant. Remarkably, the temperature readout of the sensor was well maintained despite the changes of strain and pH, indicating high selectivity towards temperature among other stimuli. This unique feature was attributed to the ratiometric detection that made the temperature readout intrinsically self-calibrated and robust to other stimuli interferences. The above results demonstrate our sensor can be used for fast and quantitative temperature monitoring with high selectivity, accuracy, and repeatability.
pH sensing performances of the sensor
Epidermal pH is an important indicator of human health and illness that can be used for medical diagnosis and health monitoring. For example, the sweat pH of a healthy human normally ranges between 4.5–6.5, whereas the sweat of patients with cystic fibrosis is usually alkaline and can have a pH value beyond 8 [29]. For pH sensing, we immobilized pH-sensitive fluorescent dyes (pHrodo Red-NHS) into the PAM coating of the HPOF via covalent bonding. The porous nature of the hydrogel matrices facilitated rapid analyte exchanges between the HPOF and aqueous surroundings (e.g., sweat) through passive diffusion. The diffusion of hydrogen ions into the hydrogel coating promotes the protonation of the rhodamine chromophore, resulting in considerably increased fluorescence of the pHrodo Red in acidic pH (Fig. 3a) [30]. Figure 3b shows the emission spectra of the sensor tested with different pH solutions. It was found that the emission of the pHrodo Red at 600 nm was notably increased at lower pH, while the emission band at 650 nm was less affected. The RET process takes place due to the spectral overlap between the UCL emission at 545 nm and the absorption spectrum of the pHrodo Red (Figure S4e). The pH response of the sensor was calibrated in the range of pH 4.0-9.5 by ratiometric dual-wavelength measurements, where the intensity ratio of the pH-sensitive emission at 600 nm and the insensitive UCL emission at 545 nm as a reference signal was calculated (Fig. 3c). Henderson-Hasselbalch equation was used for fitting of the responsive curve, which indicated an apparent pKa value of ~ 6.6, suitable for physiological pH monitoring. Besides, the sensor displayed a linear pH response with sensitivity of 27% pH− 1 over the pH range of 5.0-7.5, where a low LOD of \(\pm 0.09\) was achieved from the linear fitting (inset of Fig. 3c). Figure 3d shows the response of sensor under repeated pH tests, where the sensor exhibited a reversible and reproducible output over multiple cycles with negligible hysteresis.
Selectivity is another crucial factor for practical sweat measurements due to the presence of other interference ions in sweat. As revealed in Fig. 3e, the sensor had a high selectivity towards hydrogen ions over other dominant interference ions in sweat including K+ and Na+, due to the selective protonation mechanism. The response time of the sensor to pH was about 34 s, which was sufficiently fast for use in sweat monitoring as the sweat pH generally varied on the order of minutes [31]. To test the dynamic performance of the sensor for pH determination, the sensor was continuously treated with droplet of different pH solutions. As shown in Fig. 3f, the sensor exhibited a step-increased behavior in response to solutions of increased pH, and the sensor output was recovered after the pH was returned to the initial value, which validated the reliability of sensor for real-time pH monitoring. Furthermore, the effects of strain and temperature on the pH response of the sensor were investigated in Fig. 3g. The sensor was kept in a pH 5 buffer solution, and meanwhile tested under various strains and temperatures. It was found that the pH readout was insensitive to strain but slightly affected by temperature. The temperature-dependent effect of the pH response was attributed to the temperature dependence of the fluorescence quantum yield, defined as the ratio of emitted photons to absorbed ones [32]. The pH response could be calibrated and corrected with the temperature readout to achieve high accuracy for practical use (Figure S10).
Strain sensing performances of the sensor
Accurate detection of skin strains is critical for a wide variety of applications ranging from healthcare monitoring, sport training to human-machine interfacing [33–35]. Sensitivity and stretchability are the key parameters determining the performance of strain sensors in wearable applications, especially in healthcare monitoring. A high sensitivity is required for monitoring subtle skin deformations caused by vital signs such as artery pulse and heartbeat, which typically induce a strain less than 1% [36]. Meanwhile, a high stretchability is demanded to enable high mechanical compliance with the elastic skin. We demonstrated highly sensitive strain sensing with the stretchable multimodal sensor by virtue of the wavelength-dependent absorption characteristics of the dye molecules. Upon excitation, the UCNPs generated UCL emissions that propagated in both forward and backward directions along the HPOF (Fig. 4a, b). The backward emissions were absorbed by the dye molecules distributed in the fiber coating, leading to attenuated emissions detected at the front end. When stretched, the fiber length was increased, which prolonged the interaction length of the backward emission with the dye absorbers, resulting in enlarged absorption. Figure 4c shows the transmission spectra of the HPOFs with/without dye loading, which confirmed the absorption effect of light propagating through the HPOF in presence of dye molecules. The emission spectra of the sensor under various strains were characterized as presented in Fig. 4d. Upon stretching, the local stress deformed the coupling joint between the HPOF and the silica MMFs, and induced additional coupling loss that caused decreased intensities at all emission bands. To eliminate the absorption-independent loss effect, a dual-wavelength differential absorption method was employed, for which the emissions at 545 nm (near the absorption peak) and 655 nm (outside the absorption band) were chosen as the probing light and reference light, respectively. Figure 4e shows the differential attenuation changes at the two wavelengths versus the applied strains ranging from 0–20%. As expected, the sensor displayed increased attenuation with the increasing strain due to the increased light absorption. Notably, the sensor could sensitively detect small-scale strains below 1% with high linearity (\({{\text{R}}^2}=0.99\)), and presented a LOD as low as \(\pm 0.07\%\) (inset of Fig. 4e). The exceptional sensitivity of the sensor makes it a promising candidate to accurately detect subtle skin strains in health monitoring applications.
The temporal readout of the sensor in response to step-increased strains was further investigated (Fig. 4f). The sensor indicated an instant change of signal at each step of strain increase, and its readout was recovered once the loading strain was fully released. To evaluate the response speed, the sensor was loaded with a quasi-transient step strain of ~ 2% (Fig. 4g). The response and recovery times of the sensor were measured to be 23 ms and 25 ms, which were fast enough to capture almost all the skin strain-related physiological signals (typically < 10 Hz). Figure 4h shows the sensor readout under a cyclic loading-unloading test of different strains. The sensor exhibited a repeatable response to the sequence of loading-unloading cycles with negligible hysteresis or drift, indicating high stability and reproducibility. Moreover, the crosstalk effects of temperature and pH on the strain response of the multimodal sensor were analyzed under different temperature and pH conditions, where no strain was loaded. As shown in Fig. 4i, the sensor maintained a stable strain readout at different temperatures in the physiological range of 30–40 ℃, indicating temperature insensitivity of the strain response. In contrast, the changes of pH induced an obvious drift in the strain readout attributed to the pH-dependent absorption of the pHrodo dye. To minimize the interference of pH, the strain readout could be calibrated from the pH value determined by the multimodal sensor as the pH response was not affected by strain. Alternatively, it might be more straightforward to retrieve the strain stimuli by filtering and frequency analysis since the strain-related vital signs possessed a signal frequency much higher than that of the skin pH. In addition, long-term stability of all the sensor readouts were verified by keeping the sensor at constant strain, temperature, and pH levels, where the readout changes were monitored every 24 h for 72 h (Figure S11). No obvious drift was observed in all readouts of the sensor during the monitoring period, suggesting a high stability in long-term operation.
Real-time multiparameter monitoring of human health
Benefitting from its high stretchability, favorable biocompatibility, and multimodal sensing capabilities, the sensor could be conformally attached onto the human skin for simultaneous monitoring of multiple sets of health indicators. As a proof-of-concept demonstration, the sensor was worn on the human wrist (male, 22 years old) with sweat on the skin to detect artery pulse, skin temperature, and sweat pH in real time (Fig. 5a). Figure 5b shows the real-time skin temperature and pH value measured with our sensor. The skin temperature and sweat pH were measured to be around 33.6 ℃ and 4.8, respectively, within the normal physiological ranges. A commercially available IR sensor and pH sensor were also employed to provide reference measurements. The measured results of our sensor were in good agreement with the commercial sensors (relative error < 2%), indicating a high accuracy of the sensor for wearable temperature and pH monitoring. Moreover, artery pulse on the wrist could be simultaneously detected from the strain readout after high-pass filtering (cut-off frequency, 0.1 Hz). The wrist pulse is a weak but crucial physiological signal that can reveal the health status of both heart and arteries. Figure 5c shows the real-time artery pulse signal captured by the sensor, where each cycle of the pulse involved contraction and relaxation of the heart. The frequency of the pulse signal was observed to be 80 beats per min (bpm), in accordance with the normal range of pulse rates (60–100 bpm for healthy adults [37]). The inset of Fig. 5c shows an enlarged view of a typical pulse waveform, where three distinct peaks featured as the percussion (P) wave, tidal (T) wave, and dicrotic (D) wave could be clearly distinguished. The percentage of the T wave height divided by P wave height is defined as the augmentation index (AI), a well-established indicator of arterial stiffness and vascular aging [38]. From the measured waveform, the AI was calculated to be 0.39, which was a characteristic value expected for a healthy 22-year-old male [39].
Apart from the wrist, the sensor was further attached onto the skin of chest to detect the chest-wall movement caused by cardiopulmonary activities, and also simultaneously monitor the body temperature and sweat pH (Fig. 5d). Cardiopulmonary activity monitoring plays a critical role in evaluation of cardiac and pulmonary functions, promoting early detection and prevention of cardiovascular and pulmonary diseases such as atherosclerosis, heart failure, and asthma [40, 41]. Abnormal respiratory rate (RR) and heart rate (HR) are often the early signs of serious cardiopulmonary disorders. Figure 5e shows the original strain readout of the sensor in response to the chest movement. Periodic respiratory activity could be clearly recognized from the original signal, arising from vigorous stretching and releasing of the sensor upon exhaling and inhaling (inset of Fig. 5e). In contrast, the cardiac activity was much weaker and submerged in the respiratory signal. The two signals could be separated in the frequency domain by filtering strategy due to the different frequency bands of respiration (0.1–0.5 Hz) and cardiac beat (0.8-2Hz). Figure 5f shows the respiratory and heartbeat signals in the time domain decomposed after band-pass filtering. The obtained signals were further analyzed in the frequency domain by fast Fourier transform (FFT), which indicated a normal RR and HR of 13 bpm and 84 bpm, respectively, demonstrating great promise of our sensor for wearable cardiopulmonary monitoring in different healthcare and clinical settings. Furthermore, other vital indicators of health including skin temperature and sweat pH could be accurately detected from the calibrated temperature and pH readout. As validated in Fig. 5g, the sensor could simultaneously and continuously monitor heartbeat, respiration, skin temperature, and sweat pH without any crosstalk, and the results showed good consistency with the reference data. To our knowledge, this is the first demonstration of simultaneous monitoring of cardiopulmonary activities, body temperature, and sweat biomarker with only a single sensor.