Ultrasound-modulated optical glucose sensing using a 1645 nm laser

Regular and frequent blood glucose monitoring is vital in managing diabetes treatment plans and preventing severe complications. Because current invasive techniques impede patient compliance and are not infection-free, many noninvasive methods have been proposed. Among them, optical methods have drawn much attention for their rich optical contrast, but their resolution is degraded in deep tissue. Here, we present an ultrasound-modulated optical sensing (UOS) technique to noninvasively monitor glucose that uses an infrared laser (1645 nm) and a single-element focused ultrasound transducer. Focused ultrasound waves can acoustically localize diffused photons in scattering media, and thus optical contrast can be represented with much enhanced spatial resolution. To maximize the signal-to-noise ratio, we compared the modulation depths of UOS signals in both continuous and burst ultrasound transmission modes. Finally, UOS measurements of various glucose concentrations are presented and compared with those acquired in phantoms with a conventional diffuse optical sensing method. The UOS measurements in a 20 mm thick tissue-mimicking phantom show 26.6% accuracy in terms of mean absolute relative difference (MARD), which indicates the great potential of the proposed technique as a noninvasive glucose sensor.

In this paper, we propose an ultrasound-modulated optical sensing (UOS) system for noninvasive glucose measurement that uses an infrared laser and a single-element focused ultrasound transducer. To achieve high optical absorption contrast, we use a 1645 nm laser, based on the measured glucose absorbance. The modulation efficiencies of continuous ultrasound waves and ultrasound bursts are experimentally compared while maintaining the ultrasound intensity under the diagnostic ultrasound safety limit. Finally, using a vessel-mimicking phantom, we explore the dependency of UOS signals on glucose concentration by acquiring UOS signals from various glucose concentrations.

Results
Ultrasound-modulated optical glucose sensing system. Figure 1 shows a schematic and photograph of the UOS system. A diode-pumped solid state laser (MIL-N-1645, CNI Laser, China) with a wavelength of 1645 nm and a coherence length of > 10 cm irradiated a tissue-mimicking phantom. The laser fluence on the target surface was about 40 mJ/cm 2 for 5 ms per each measurement, which is far below the ANSI safety standard of 1,489 mJ/cm 2 for 1645 nm laser exposure on skin 43 . Ultrasound waves were generated and focused into the phantom by an ultrasound transducer (KPS100-1-P38, The Ultran Group, USA) with a central frequency of 1 MHz, an active diameter of 25 mm, and a focal length of 38 mm. The ultrasound and optical propagation axes were set perpendicular to each other, and the ultrasound focus was located at the intersection of the two paths. A function generator (33220A, Agilent Technologies, USA) produced 1 MHz sinusoidal waveforms in continuous wave (CW) or burst mode, which were then amplified by an RF power amplifier (325LA, Electronics and Innovation, USA) with a fixed gain of 50 dB to drive the ultrasound transducer. The light transmitted through the phantom was collected by a photo diode (PDA10D2, Thorlabs, USA; ø1.0 mm photosensitive area; 900-2,600 nm spectral response range). The corresponding light intensity was recorded by a data acquisition system (DAQ; ATS9350, Alazar Technologies, Canada; 12 bit resolution; up to 500 MS/s sampling rate; ± 4 V input range). The data acquisition was synchronized with the ultrasound firing using the trigger signal from the function generator. A UV fused silica beam splitter (BSN12R, Thorlabs, USA; 10:90 reflectance-to-transmission split ratio) split the input light, which was collected by a power meter (G8931-20, Hamamatsu Photonics, Japan; ø0.2 mm photosensitive area; 950-1,700 nm spectral response range) to normalize the light fluence.
optical properties of the in vitro phantom. We evaluated the optical properties of the in vitro phantom to validate its similarity to biological tissues. The background of the phantom was composed of 0.2 g/mL gelatin, 0.1 mg/mL TiO 2 powder (232033, Sigma-Aldrich, USA), and heavy water (151882, Sigma-Aldrich, USA), with dimensions of 60 mm (X) × 20 mm (Y) × 40 mm (Z) 44 . The measured effective attenuation coefficient was 3.1 cm -1 at 1645 nm. To prove the similarity between blood vessel and the tubing used for mimicking blood vessel, we excised carotid artery of a rat (female, 8 weeks) and measured the attenuation of light in the vessel while sandwiching it between two sheets of cover glass (Cover Slips, Duran Group, Germany; 24 mm × 50 mm, www.nature.com/scientificreports/ 0.16-0.19 mm thickness). The size of the excised blood vessel was about 7.5 mm × 2.5 mm. All of the animal procedures were conducted in accordance with the Pohang University of Science and Technology (POSTECH) Institutional Animal Care and Use Committee (IACUC) protocols. We also measured the attenuation of light in the tubing in the same manner. The transmitted laser power after the cover glass, the cover glass embedding the blood vessel, and the cover glass embedding the tubing were 93.9%, 87.2%, and 86.1%, respectively, which shows their similarity in optical attenuation.
optical absorption contrast of glucose solution. To verify the optical absorption contrast of glucose solution, we separately obtained the absorbance spectra of various concentrations of glucose solutions and heavy water. Figure 2a and b show the optical transmittance and the absorbance of the glucose solutions and heavy water. Here, 0 mg/dL denotes the heavy water. The absorbance beyond 1,850 nm is omitted because converted absorbance values in a range where the transmittance is close to zero have low reliability. At 1645 nm, the optical wavelength used in the developed UOS system (gray dotted lines in Fig. 2a,b), the absorbance distinctly increases with concentration of glucose solution while having relatively high water transmittance of 54%, which could achieve a high glucose absorption contrast over water. Figure 2c represents the absorbance values of the glucose solutions at 1645 nm and the fitted linear regression line. The measured absorbance at 1645 nm exhibits direct linear response to the concentration of glucose solution with an R-squared value of 0.9994, which demonstrates the capability for glucose sensing using the proposed optical wavelength in water-based medium such as biological tissues.

Modulation depth enhancement with acoustic bursts. The international standard IEC 60601-2-37
and the FDA 45,46 stipulate that for medical diagnostic ultrasound systems in non-ophthalmic uses the thermal and mechanical indices should be less than 6.0 and 1.9, respectively. IEC 62,359 defines the mechanical index (MI) and the soft-tissue thermal index (TIS) as where z is the distance from the external transducer aperture to the point of interest; p r.3 (z) , W .3 (z) , and I SPTA.3 (z) are the attenuated peak rarefaction pressure, power, and spatial-peak temporal-average intensity at a depth z with an acoustic attenuation coefficient of 0.3 dB/cm/MHz; f awf is the acoustic working frequency; and C MI , C TIS1 , and C TIS2 are constants of 1 MPa·MHz -1/2 , 210 mW·MHz, and 210 mW·cm −2 ·MHz, respectively [47][48][49] . By definition, when applied the same driving voltage, MIs in CW and burst modes would be the same and TIS in burst mode would be lower in proportion to the duty cycle α ( 0 ≤ α ≤ 1 ) than TIS in CW mode. To find the maximum driving voltage satisfying both the MI and TIS safety limits, we measured MI and TIS values by applying various driving voltages to the transducer in CW and burst modes, using an acoustic intensity measurement system (AIMS III, Onda, USA). In both CW and burst modes, 1 MHz sinusoidal waves were applied, and in burst mode, 10 μs bursts with a repetition period of 100 μs, i.e. a duty cycle of 0.1, were applied. Table 1 shows the maximum driving voltages within MI (V MAX, MI ), and TIS (V MAX, TIS ). The maximum allowable driving voltage within the MI limit for CW mode and that for burst mode are almost the same, whereas the maximum allowable driving voltage within the TIS limit for burst mode is about 3.3 times higher than that for CW mode, as expected 50,51 . Note that the theoretical value of the ratio between V MAX, TIS in burst mode with a duty cycle of 0.1 and that in CW mode is 1 Using the maximum allowable driving voltages within the diagnostic ultrasound safety limits (MI < 1.9 and TIS < 6.0), we obtained UOS signals in CW and burst modes from a 20-mm-thick tissue-mimicking phantom. Figure 3 shows the resultant time-resolved UOS signals in CW and burst modes. The light intensity is modulated at the ultrasound frequency during the entire time window in CW mode, and during the duty cycle (0.1 for a 100 μs period = 10 μs) in burst mode. The modulated light starts to appear at about 27 μs in burst mode, which matches well with the ultrasound time of flight to the crossing point of the light and ultrasound paths at the focal length (38 mm) of the ultrasound transducer, which can be converted to 25.3 μs assuming the speed of sound is 1,500 m/s. The modulation depth, defined as the ratio of the ultrasound-modulated light intensity to the unmodulated light intensity, was obtained as 0.062 and 0.154 in CW and burst modes, respectively ( Table 1). As expected, we could achieve about 2.5 times greater modulation depth in burst mode than in CW mode. This greater depth was possible because we could apply much higher acoustic pressure to the sample in burst mode, since the sample could cool during the off cycle, whereas the sample remained heated in CW mode.
in vitro vessel-mimicking phantom measurements. To demonstrate the feasibility of glucose sensing using the developed UOS system, we obtained UOS signals from various glucose concentrations in a vessel-mimicking phantom. Ultrasound bursts with 10% duty cycle were used to achieve high SNR and prevent the phantom from heating up. Figure 4a shows the normalized modulation depths and diffuse transmittances extracted from the UOS and the conventional diffuse optical sensing results, respectively, and Fig. 4c is a closeup of the measurements within a range of 0-400 mg/dL. The error bars represent ± 1 standard deviation of 20 measurements. The dashed lines and the shaded areas indicate curves of normalized signal fitted to glucose concentration and their 95% confidence intervals. We can clearly see that modulation depth decreases dramatically as glucose concentration increases, but there is no noticeable dependency between diffuse transmittance and glucose concentration in the biologically relevant range (Fig. 4c). Figure 4b shows the predicted glucose concentration of an additional five measurements for each glucose concentration, using the calibration result in Fig. 4a. Figure 4d shows the Clarke's error grid analysis 52 of the predicted glucose concentration obtained from the additional five UOS measurements for each glucose concentration, and the calibration results are in Fig. 4c. The prediction accuracy is assessed as 26.6% mean absolute relative difference (MARD), defined by the arithmetic mean of relative absolute differences as follows 53-55 : where N is the total number of measurements and y est,i and y ref ,i are the estimated and reference values at the i-th measurement, respectively. In the assessment, only data in the range 0-400 mg/dL are used. Note that the

Discussion
This study has demonstrated the feasibility of a UOS glucose meter in a vessel-mimicking phantom. To achieve strong glucose absorption contrast, we used a 1645 nm laser based on the measured absorbance spectra. This wavelength lies in the first overtone band (1,500-1,800 nm) known to have primary absorption peaks of glucose in the NIR region 56,57 . Two different ultrasound transmission modes, bursts with a 10% duty cycle and continuous waves, were used to modulate diffusive photons in a 20-mm-thick phantom, and the resultant modulation depths were compared. The sensing region was targeted at the middle of the 20-mm-thick phantom, i.e. at 10 mm depth, taking into account the typical depth of the subcutaneous tissues (2-30 mm) where blood glucose is desired to be measured 58,59 . Within the diagnostic ultrasound safety limit, modulation depth in burst mode was improved by 2.5 times over CW mode. We could apply much higher acoustic pressure to the sample in burst mode since www.nature.com/scientificreports/ the sample could cool during the off cycle, whereas it remained heated in CW mode. The predicted glucose concentration from UOS measurements had a 26.6% MARD accuracy, which demonstrates the feasible use of the device as a noninvasive glucose meter. The measured absorbance values of glucose solution and heavy water at 1645 nm were 0.523 and 0.264, respectively, which can be converted to absorption coefficients of 1.20 cm −1 and 0.61 cm −1 , respectively. The absorption coefficients of oxyhemoglobin, deoxyhemoglobin, and lipid at 1,600 nm are about 0.9 cm −1 , 0.8 cm −1 , and 0.25 cm −1 , respectively 60 , all of which are lower than the measured absorption coefficient of glucose solution. In the wavelength region of interest in this study, optical absorption in biological tissues is mainly attributed to water 60-64 . In this regard, the feasibility of the proposed technique as a blood glucose sensor was first demonstrated based on the absorbance of glucose solution in water higher than the absorbance of water. Indeed, the absorbance of glucose dissolved in deionized water (0.3 [a.u.]) was higher than that of deionized water (0.06 [a.u.]) at 1645 nm 65 . Even the absorbance of blood at 1645 nm increased as blood glucose level increased within the biologically relevant range 66 .
The effective attenuation coefficient of the phantom was measured as 3.1 cm −1 , which is slightly lower than that of human tissue, about 12 cm −1 at 1645 nm 67,68 . To assess whether the proposed system would be feasible in humans, we estimated the expected maximum penetration depth in human tissue. Assume that the minimum detectable power ( P min ) of the proposed system is three times NEP (noise equivalent power) of the photodiode, which is calculated as With the output power ( P 0 ) of the laser, 250 mW, the power ( P ) after penetrating human tissue with a depth of z and the maximum penetration depth ( z MAX ) can be calculated as follows.
The maximum penetration depth in human tissue is estimated as 0.54 cm, which is suitable for our target depth. This could be more increased by using a high-sensitive photodetector (e.g. avalanche photodiode) or a high power laser. Note that we used laser power less than 3% of ANSI safety limit in this study.
Thanks to the fact that ultrasound scatters much less than light does in biological tissue, diffusive photons could be localized by ultrasonic modulation and optical properties could be sensed at much higher spatial resolution than with conventional pure optical sensing techniques. Taking this advantage, the proposed technique could be a promising candidate for sensing blood glucose levels in a deep dermal vascular plexus rather than sensing interstitial fluid glucose levels in superficial layers which most noninvasive glucose monitoring techniques are targeted at. Photoacoustic sensing also can provide similar advantages, but requires an expensive and bulky pulsed laser, whereas the proposed UOS system utilizes a simple CW laser and a single element ultrasound transducer, which can easily be miniaturized at low cost and be an excellent candidate for a noninvasive homecare monitoring device.
For successful in vivo application, several challenges are needed to be overcome. (1) This preliminary study implemented the UOS system in transmission mode, which detects diffuse transmittance through scattering phantoms. By modifying the system to work in reflection mode, the technique could be extended to monitoring blood glucose levels in vivo. (2) Pulsatile movement might affect UOS measurements. Since UOS acquisition time in this study (5 ms) is significantly shorter than pulsation repetition period (0.6-1.0 s), we can reduce, or even eliminate the effect by synchronizing UOS measurements with pulsation repetition period. ECG signal can be used for synchronization, or ultrasound itself can be used by exploiting Doppler ultrasonography. (3) Absorption of aqueous glucose could be affected by temperature at the site of measurement. Using the calibration curve for changes in absorbance over temperature 69 , the error due to temperature dependence could be corrected. Thus, further studies incorporating ergonomic design, human blood sample tests, in vivo trials, and calibration for individual subject would be desirable. Even though our study is limited to phantom, we believe that our results proved the feasibility of the proposed technique as a noninvasive glucose meter.

Methods optical absorption contrast of glucose solution.
A UV-Vis-NIR spectrophotometer (Cary 5000, Varian, Australia) was used to obtain the absorbance spectra of glucose solutions at various concentrations. The highest concentration of the glucose solutions was 9,333 mg/dL, achieved by dissolving dextrose powder (D9434, Sigma-Aldrich, USA) in heavy water, and was diluted to various concentrations down to 1% (93 mg/ dL). The absorbance spectrum of heavy water, denoted by 0 mg/dL, was also obtained as a control. Each sample was separately placed in a quartz cuvette with a 1 cm light path ( l ), and their transmittance was measured over a wavelength range of 500-2,000 nm, with a 1 nm step size. The absorbance ( A ) was converted from the measured percent transmittance ( %T ) using the equation A = 2 − log(%T) , and absorption coefficient can be calculated as µ a = ln(10) × A/l. in vitro phantom preparation. A mold was custom-designed with dimensions of 60 mm (X) × 20 mm (Y) × 40 mm (Z) using 1-mm-thick acryl. A lumen tubing (72D Pebax Tubing, Duke Extrusion, USA; 0.156 inch www.nature.com/scientificreports/ outer diameter; 0.005 inch nominal wall thickness) was positioned in the middle of the phantom mold with respect to the light propagation path (Y) and parallel to the X-Z plane (Fig. 5a). The overlap between the diffused light zone (red-filled area in Fig. 5a) and the ultrasound propagation zone (green-filled area in Fig. 5a) was the localized sensing region. The overlap zone measured about ø4 mm (X-Y plane) × 15 mm (Z) (magenta-filled area in Fig. 5b). To mimic soft tissues, 0.1 mg/mL TiO 2 powder (232033, Sigma-Aldrich, USA) was dispersed in heavy water (151882, Sigma-Aldrich, USA) and then 0.2 g/mL gelatin was dissolved in the mixture. The gelatin-TiO 2 mixture was poured into the mold and solidified with an acrylic lid closed. To increase the efficiency of ultrasound delivery, an acoustic standoff was designed using 3D design software (Inventor, Autodesk, USA) and printed using a 3D printer (ProJet MJP 2500, 3D Systems, USA). The top and the bottom of the acoustic standoff were covered with a thin polycarbonate film, and the inside was filled with water for acoustic coupling.
in vitro vessel-mimicking phantom measurements. To minimize errors, we first prepared glucose solution with a concentration of 10,000 mg/dL and then diluted the solution in 1,000 mg/dL intervals from 1,000 to 9,000 mg/dL, and in 50 mg/dL intervals from 0 to 400 mg/dL. Each solution was then injected into the tubing (blue dotted area in Fig. 5b) and its UOS signal was measured with an optical wavelength of 1,645 nm. We used 10 ultrasound bursts with a carrier frequency of 1 MHz and a repetition period of 100 μs. The signals were averaged over 50 periods (100 μs/period × 50 periods = 5 ms).